Merge branch 'master' of github.com:ludv1x/opencv_contrib

modules/dnn/CMakeLists.txt

@@ -33,9 +33,18 @@ ocv_module_include_directories(include src/caffe ${PROTOBUF_INCLUDE_DIR})
 
 ocv_create_module(${PROTOBUF_LIBRARIES})
 
+ocv_add_samples()
 ocv_add_accuracy_tests()
 ocv_add_perf_tests()
-ocv_add_samples()
+
+OCV_OPTION(${the_module}_DOWNLOAD_CAFFE_MODELS "Use GoogLeNet Caffe model for testing" ON IF BUILD_TESTS AND PYTHON2_EXECUTABLE AND DEFINED ENV{OPENCV_TEST_DATA_PATH})
+if(BUILD_TESTS AND ${the_module}_DOWNLOAD_CAFFE_MODELS)
+    add_custom_command( TARGET opencv_test_${name} POST_BUILD
+                        COMMAND ${PYTHON2_EXECUTABLE} download_model.py test_models.json
+                        WORKING_DIRECTORY ${CMAKE_CURRENT_SOURCE_DIR}/scripts )
+else()
+    add_definitions(-DDISABLE_CAFFE_MODEL_TESTS=1)
+endif()
 
 else()#build as standalone module (for development purposes)
 

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

@@ -0,0 +1,142 @@
+#ifndef __OPENCV_DNN_DNN_BLOB_HPP__
+#define __OPENCV_DNN_DNN_BLOB_HPP__
+#include <opencv2/core.hpp>
+#include <vector>
+#include <ostream>
+
+namespace cv
+{
+namespace dnn
+{
+    struct BlobShape
+    {
+        explicit BlobShape(int ndims = 4, int fill = 1);
+        BlobShape(int num, int cn, int rows, int cols);
+        BlobShape(int ndims, const int *sizes);
+        BlobShape(const std::vector<int> &sizes);
+        template<int n>
+        BlobShape(const Vec<int, n> &shape);
+
+        int dims() const;
+        int size(int axis) const;
+        int &size(int axis);
+
+        //do the same as size()
+        int operator[](int axis) const;
+        int &operator[](int axis);
+
+        //same as size(), but size of non-existing dimensions equal to 1
+        int xsize(int axis) const;
+
+        ptrdiff_t total();
+
+        const int *ptr() const;
+
+        bool equal(const BlobShape &other) const;
+
+    private:
+        cv::AutoBuffer<int,4> sz;
+    };
+
+    bool operator== (const BlobShape &l, const BlobShape &r);
+
+    //maybe useless
+    CV_EXPORTS std::ostream &operator<< (std::ostream &stream, const BlobShape &shape);
+
+
+    /** @brief provides convenient methods for continuous n-dimensional array processing, dedicated for convolution neural networks
+    It's realized as wrapper over \ref cv::Mat and \ref cv::UMat and will support methods for CPU/GPU switching
+    */
+    class CV_EXPORTS Blob
+    {
+    public:
+        explicit Blob();
+        /** @brief constucts 4-dimensional blob from input
+         *  @param in 2-dimensional or 3-dimensional single-channel image (or vector from them)
+         *  @param dstCn if specified force size of ouptut blob channel-dimension
+        */
+        explicit Blob(InputArray in, int dstCn = -1);
+
+        void create(const BlobShape &shape, int type = CV_32F);
+
+        void fill(InputArray in);
+        void fill(const BlobShape &shape, int type, void *data, bool deepCopy = true);
+
+        Mat& getMatRef();
+        const Mat& getMatRef() const;
+        Mat getMat(int n, int cn);
+
+        //shape getters
+        ///returns real count of blob dimensions
+        int dims() const;
+
+        /** @brief returns size of corresponding dimension (axis)
+        @param axis dimension index
+        Python-like indexing is supported, so \p axis can be negative, i. e. -1 is last dimension.
+        Supposed that size of non-existing dimensions equal to 1, so the method always finished.
+        */
+        int xsize(int axis) const;
+
+        /** @brief returns size of corresponding dimension (axis)
+        @param axis dimension index
+        Python-like indexing is supported, so \p axis can be negative, i. e. -1 is last dimension.
+        @note Unlike ::xsize, if \p axis points to non-existing dimension then an error will be generated.
+        */
+        int size(int axis) const;
+
+        /** @brief returns number of elements
+        @param startAxis starting axis (inverse indexing can be used)
+        @param endAxis ending (excluded) axis
+        @see ::canonicalAxis
+        */
+        size_t total(int startAxis = 0, int endAxis = -1) const;
+
+        /** @brief converts axis index to canonical format (where 0 <= axis <= ::dims)
+        */
+        int canonicalAxis(int axis) const;
+
+        /** @brief returns real shape of the blob
+        */
+        BlobShape shape() const;
+
+        bool equalShape(const Blob &other) const;
+
+        //shape getters, oriented for 4-dim Blobs processing
+        int cols() const;
+        int rows() const;
+        int channels() const;
+        int num() const;
+        Size size2() const;
+        Vec4i shape4() const;
+
+        //CPU data pointer functions
+        int offset(int n = 0, int cn = 0, int row = 0, int col = 0) const;
+        uchar *ptrRaw(int n = 0, int cn = 0, int row = 0, int col = 0);
+        float *ptrf(int n = 0, int cn = 0, int row = 0, int col = 0);
+        template<typename TFloat>
+        TFloat *ptr(int n = 0, int cn = 0, int row = 0, int col = 0);
+
+        /** @brief share data with other blob and returns *this
+        @returns *this
+        */
+        Blob &shareFrom(const Blob &blob);
+        /** @brief adjust blob shape to required (data reallocated if needed)
+        @returns *this
+        */
+        Blob &reshape(const BlobShape &shape);
+
+        int type() const;
+        bool isFloat() const;
+        bool isDouble() const;
+
+    private:
+        const int *sizes() const;
+
+        Mat m;
+    };
+}
+}
+
+#include "blob.inl.hpp"
+
+#endif

modules/dnn/include/opencv2/dnn/blob.inl.hpp

@@ -0,0 +1,284 @@
+#ifndef __OPENCV_DNN_DNN_BLOB_INL_HPP__
+#define __OPENCV_DNN_DNN_BLOB_INL_HPP__
+#include "blob.hpp"
+
+namespace cv
+{
+namespace dnn
+{
+
+inline BlobShape::BlobShape(int ndims, int fill) : sz( (size_t)std::max(ndims, 1) )
+{
+    for (int i = 0; i < ndims; i++)
+        sz[i] = fill;
+}
+
+inline BlobShape::BlobShape(int ndims, const int *sizes) : sz( (size_t)std::max(ndims, 1) )
+{
+    CV_Assert(ndims > 0);
+    for (int i = 0; i < ndims; i++)
+        sz[i] = sizes[i];
+}
+
+inline BlobShape::BlobShape(int num, int cn, int rows, int cols) : sz(4)
+{
+    sz[0] = num;
+    sz[1] = cn;
+    sz[2] = rows;
+    sz[3] = cols;
+}
+
+inline BlobShape::BlobShape(const std::vector<int> &sizes) : sz( sizes.size() )
+{
+    CV_Assert(sizes.size() > 0);
+    for (int i = 0; i < (int)sizes.size(); i++)
+        sz[i] = sizes[i];
+}
+
+template<int n>
+inline BlobShape::BlobShape(const Vec<int, n> &shape) : sz(n)
+{
+    for (int i = 0; i < n; i++)
+        sz[i] = shape[i];
+}
+
+inline int BlobShape::dims() const
+{
+    return (int)sz.size();
+}
+
+inline int BlobShape::xsize(int axis) const
+{
+    if (axis < -dims() || axis >= dims())
+        return 1;
+
+    return sz[(axis < 0) ? axis + dims() : axis];
+}
+
+inline int BlobShape::size(int axis) const
+{
+    CV_Assert(-dims() <= axis && axis < dims());
+    return sz[(axis < 0) ? axis + dims() : axis];
+}
+
+inline int &BlobShape::size(int axis)
+{
+    CV_Assert(-dims() <= axis && axis < dims());
+    return sz[(axis < 0) ? axis + dims() : axis];
+}
+
+inline int BlobShape::operator[] (int axis) const
+{
+    CV_Assert(-dims() <= axis && axis < dims());
+    return sz[(axis < 0) ? axis + dims() : axis];
+}
+
+inline int &BlobShape::operator[] (int axis)
+{
+    CV_Assert(-dims() <= axis && axis < dims());
+    return sz[(axis < 0) ? axis + dims() : axis];
+}
+
+inline ptrdiff_t BlobShape::total()
+{
+    CV_Assert(dims() >= 1);
+
+    ptrdiff_t res = 1;
+    for (int i = 0; i < dims(); i++)
+        res *= sz[i];
+    return res;
+}
+
+
+inline const int *BlobShape::ptr() const
+{
+    return sz;
+}
+
+inline bool BlobShape::equal(const BlobShape &other) const
+{
+    if (this->dims() != other.dims())
+        return false;
+
+    for (int i = 0; i < other.dims(); i++)
+    {
+        if (sz[i] != other.sz[i])
+            return false;
+    }
+
+    return true;
+}
+
+inline bool operator== (const BlobShape &l, const BlobShape &r)
+{
+    return l.equal(r);
+}
+
+
+
+inline int Blob::canonicalAxis(int axis) const
+{
+    CV_Assert(-dims() <= axis && axis < dims());
+
+    if (axis < 0)
+    {
+        return dims() + axis;
+    }
+    return axis;
+}
+
+inline int Blob::dims() const
+{
+    return m.dims;
+}
+
+inline int Blob::xsize(int axis) const
+{
+    if (axis < -dims() || axis >= dims())
+        return 1;
+
+    return sizes()[(axis < 0) ? axis + dims() : axis];
+}
+
+inline int Blob::size(int axis) const
+{
+    CV_Assert(-dims() <= axis && axis < dims());
+    return sizes()[(axis < 0) ? axis + dims() : axis];
+}
+
+inline size_t Blob::total(int startAxis, int endAxis) const
+{
+    if (startAxis < 0)
+        startAxis += dims();
+
+    if (endAxis == -1)
+        endAxis = dims();
+
+    CV_Assert(0 <= startAxis && startAxis <= endAxis && endAxis <= dims());
+
+    size_t size = 1; //assume that blob isn't empty
+    for (int i = startAxis; i < endAxis; i++)
+        size *= (size_t)sizes()[i];
+
+    return size;
+}
+
+inline int Blob::offset(int n, int cn, int row, int col) const
+{
+    CV_DbgAssert(0 <= n && n < num() && 0 <= cn && cn < channels() && 0 <= row && row < rows() && 0 <= col && col < cols());
+    return ((n*channels() + cn)*rows() + row)*cols() + col;
+}
+
+inline float *Blob::ptrf(int n, int cn, int row, int col)
+{
+    CV_Assert(type() == CV_32F);
+    return (float*)m.data + offset(n, cn, row, col);
+}
+
+inline uchar *Blob::ptrRaw(int n, int cn, int row, int col)
+{
+    return m.data + m.elemSize() * offset(n, cn, row, col);
+}
+
+template<typename TFloat>
+inline TFloat* Blob::ptr(int n, int cn, int row, int col)
+{
+    CV_Assert(type() == cv::DataDepth<TFloat>::value);
+    return (TFloat*) ptrRaw(n, cn, row, col);
+}
+
+inline BlobShape Blob::shape() const
+{
+    return BlobShape(dims(), sizes());
+}
+
+inline bool Blob::equalShape(const Blob &other) const
+{
+    if (this->dims() != other.dims())
+        return false;
+
+    for (int i = 0; i < dims(); i++)
+    {
+        if (this->sizes()[i] != other.sizes()[i])
+            return false;
+    }
+    return true;
+}
+
+inline Mat& Blob::getMatRef()
+{
+    return m;
+}
+
+inline const Mat& Blob::getMatRef() const
+{
+    return m;
+}
+
+inline Mat Blob::getMat(int n, int cn)
+{
+    return Mat(rows(), cols(), m.type(), this->ptrRaw(n, cn));
+}
+
+inline int Blob::cols() const
+{
+    return xsize(3);
+}
+
+inline int Blob::rows() const
+{
+    return xsize(2);
+}
+
+inline int Blob::channels() const
+{
+    return xsize(1);
+}
+
+inline int Blob::num() const
+{
+    return xsize(0);
+}
+
+inline Size Blob::size2() const
+{
+    return Size(cols(), rows());
+}
+
+inline int Blob::type() const
+{
+    return m.depth();
+}
+
+inline bool Blob::isFloat() const
+{
+    return (type() == CV_32F);
+}
+
+inline bool Blob::isDouble() const
+{
+    return (type() == CV_32F);
+}
+
+inline const int * Blob::sizes() const
+{
+    return &m.size[0];
+}
+
+
+inline Blob &Blob::shareFrom(const Blob &blob)
+{
+    this->m = blob.m;
+    return *this;
+}
+
+inline Blob &Blob::reshape(const BlobShape &shape)
+{
+    m = m.reshape(1, shape.dims(), shape.ptr());
+    return *this;
+}
+
+}
+}
+
+#endif

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

@@ -1,5 +1,5 @@
-#ifndef __OPENCV_DNN_DICT_HPP__
-#define __OPENCV_DNN_DICT_HPP__
+#ifndef __OPENCV_DNN_DNN_DICT_HPP__
+#define __OPENCV_DNN_DNN_DICT_HPP__
 
 #include <opencv2/core.hpp>
 #include <map>
@@ -11,41 +11,77 @@ namespace dnn
 
 struct DictValue
 {
-    int type;
-
-    union
-    {
-        int64 i;
-        double d;
-        bool b;
-        String *s;
-    };
-
     DictValue(const DictValue &r);
-    DictValue(int p = 0)        : type(cv::Param::INT), i(p) {}
-    DictValue(unsigned p)       : type(cv::Param::INT), i(p) {}
-    DictValue(double p)         : type(cv::Param::REAL), d(p) {}
-    DictValue(bool p)           : type(cv::Param::BOOLEAN), b(p) {}
-    DictValue(const String &p)  : type(cv::Param::STRING), s(new String(p)) {}
-    DictValue(const char *str)  : type(cv::Param::STRING), s(new String(str)) {}
+    DictValue(int p = 0)        : type(Param::INT), pi(new AutoBuffer<int64,1>) { (*pi)[0] = p; }
+    DictValue(unsigned p)       : type(Param::INT), pi(new AutoBuffer<int64,1>) { (*pi)[0] = p; }
+    DictValue(double p)         : type(Param::REAL), pd(new AutoBuffer<double,1>) { (*pd)[0] = p; }
+    DictValue(const String &p)  : type(Param::STRING), ps(new AutoBuffer<String,1>) { (*ps)[0] = p; }
+
+    template<typename TypeIter>
+    static DictValue arrayInt(TypeIter begin, int size);
+    template<typename TypeIter>
+    static DictValue arrayReal(TypeIter begin, int size);
+    template<typename TypeIter>
+    static DictValue arrayString(TypeIter begin, int size);
 
     template<typename T>
-    T get() const;
+    T get(int idx = -1) const;
+
+    int size() const;
 
-    bool isString() const;
     bool isInt() const;
+    bool isString() const;
+    bool isReal() const;
 
     DictValue &operator=(const DictValue &r);
 
     ~DictValue();
 
-private:
+protected:
+
+    int type;
+
+    union
+    {
+        AutoBuffer<int64, 1> *pi;
+        AutoBuffer<double, 1> *pd;
+        AutoBuffer<String, 1> *ps;
+        void *p;
+    };
+
+    DictValue(int _type, void *_p) : type(_type), p(_p) {}
     void release();
 };
 
+template<typename TypeIter>
+DictValue DictValue::arrayInt(TypeIter begin, int size)
+{
+    DictValue res(Param::INT, new AutoBuffer<int64, 1>(size));
+    for (int j = 0; j < size; begin++, j++)
+        (*res.pi)[j] = *begin;
+    return res;
+}
+
+template<typename TypeIter>
+DictValue DictValue::arrayReal(TypeIter begin, int size)
+{
+    DictValue res(Param::REAL, new AutoBuffer<double, 1>(size));
+    for (int j = 0; j < size; begin++, j++)
+        (*res.pd)[j] = *begin;
+    return res;
+}
+
+template<typename TypeIter>
+DictValue DictValue::arrayString(TypeIter begin, int size)
+{
+    DictValue res(Param::STRING, new AutoBuffer<String, 1>(size));
+    for (int j = 0; j < size; begin++, j++)
+        (*res.ps)[j] = *begin;
+    return res;
+}
+
 class CV_EXPORTS Dict
 {
-    //TODO: maybe this mechanism was realized somewhere in OpenCV?
     typedef std::map<String, DictValue> _Dict;
     _Dict dict;
 
@@ -62,13 +98,18 @@ public:
         return (i == dict.end()) ? NULL : &i->second;
     }
 
-    template <typename T>
-    T get(const String &name) const
+    const DictValue &get(const String &name) const
     {
         _Dict::const_iterator i = dict.find(name);
         if (i == dict.end())
-            CV_Error(cv::Error::StsBadArg, "Required argument \"" + name + "\" not found into dictionary");
-        return i->second.get<T>();
+            CV_Error(Error::StsBadArg, "Required argument \"" + name + "\" not found into dictionary");
+        return i->second;
+    }
+
+    template <typename T>
+    T get(const String &name) const
+    {
+        return this->get(name).get<T>();
     }
 
     template <typename T>
@@ -94,92 +135,89 @@ public:
 
         return value;
     }
-
-    inline void print()
-    {
-        for (_Dict::const_iterator i = dict.begin(); i != dict.end(); i++)
-        {
-            std::cout << i->first << std::endl;
-        }
-    }
 };
 
+template<>
+inline DictValue DictValue::get<DictValue>(int idx) const
+{
+    CV_Assert(idx == -1);
+    return *this;
+}
 
 template<>
-inline int DictValue::get<int>() const
+inline int64 DictValue::get<int64>(int idx) const
 {
-    CV_Assert(type == cv::Param::INT);
-    return (int)i;
+    CV_Assert(isInt());
+    CV_Assert(idx == -1 && pi->size() == 1 || idx >= 0 && idx < (int)pi->size());
+    return (*pi)[(idx == -1) ? 0 : idx];
 }
 
 template<>
-inline unsigned DictValue::get<unsigned>() const
+inline int DictValue::get<int>(int idx) const
 {
-    CV_Assert(type == cv::Param::INT);
-    return (unsigned)i;
+    return (int)get<int64>(idx);
 }
 
 template<>
-inline double DictValue::get<double>() const
+inline unsigned DictValue::get<unsigned>(int idx) const
 {
-    if (type == cv::Param::REAL)
-        return d;
-    else if (type == cv::Param::INT)
-        return (double)i;
-    else
-    {
-        CV_Assert(type == cv::Param::REAL || type == cv::Param::INT);
-        return 0;
-    }
+    return (unsigned)get<int64>(idx);
 }
 
 template<>
-inline float DictValue::get<float>() const
+inline bool DictValue::get<bool>(int idx) const
 {
-    if (type == cv::Param::FLOAT)
-        return (float)d;
-    else if (type == cv::Param::INT)
-        return (float)i;
-    else
-    {
-        CV_Assert(type == cv::Param::FLOAT || type == cv::Param::INT);
-        return (float)0;
-    }
+    return (get<int64>(idx) != 0);
 }
 
 template<>
-inline bool DictValue::get<bool>() const
+inline double DictValue::get<double>(int idx) const
 {
-    if (type == cv::Param::BOOLEAN)
+    if (type == Param::REAL)
     {
-        return b;
+        CV_Assert(idx == -1 && pd->size() == 1 || idx >= 0 && idx < (int)pd->size());
+        return (*pd)[0];
     }
-    else if (type == cv::Param::INT)
+    else if (type == Param::INT)
     {
-        return i != 0;
+        CV_Assert(idx == -1 && pi->size() == 1 || idx >= 0 && idx < (int)pi->size());
+        return (double)(*pi)[0];;
     }
     else
     {
-        CV_Assert(type == cv::Param::BOOLEAN || type == cv::Param::INT);
+        CV_Assert(isReal());
         return 0;
     }
 }
 
 template<>
-inline String DictValue::get<String>() const
+inline float DictValue::get<float>(int idx) const
 {
-    CV_Assert(type == cv::Param::STRING);
-    return *s;
+    return (float)get<double>(idx);
+}
+
+template<>
+inline String DictValue::get<String>(int idx) const
+{
+    CV_Assert(isString());
+    CV_Assert(idx == -1 && ps->size() == 1 || idx >= 0 && idx < (int)ps->size());
+    return (*ps)[(idx == -1) ? 0 : idx];
 }
 
 inline void DictValue::release()
 {
-    if (type == cv::Param::STRING && s != NULL)
+    switch (type)
     {
-        delete s;
-        s = NULL;
+    case Param::INT:
+        delete pi;
+        break;
+    case Param::STRING:
+        delete ps;
+        break;
+    case Param::REAL:
+        delete pd;
+        break;
     }
-
 }
 
 inline DictValue::~DictValue()
@@ -192,33 +230,73 @@ inline DictValue & DictValue::operator=(const DictValue &r)
     if (&r == this)
         return *this;
 
+    if (r.type == Param::INT)
+    {
+        AutoBuffer<int64, 1> *tmp = new AutoBuffer<int64, 1>(*r.pi);
         release();
-
-    //how to copy anonymous union without memcpy?
-    for (size_t i = 0; i < sizeof(*this); i++)
-        ((uchar*)this)[i] = ((uchar*)&r)[i];
-
-    if (r.type == cv::Param::STRING)
+        pi = tmp;
+    }
+    else if (r.type == Param::STRING)
     {
-        s = new String(*r.s);
+        AutoBuffer<String, 1> *tmp = new AutoBuffer<String, 1>(*r.ps);
+        release();
+        ps = tmp;
     }
+    else if (r.type == Param::REAL)
+    {
+        AutoBuffer<double, 1> *tmp = new AutoBuffer<double, 1>(*r.pd);
+        release();
+        pd = tmp;
+    }
+
+    type = r.type;
 
     return *this;
 }
 
 inline DictValue::DictValue(const DictValue &r)
 {
-    *this = r;
+    type = r.type;
+
+    if (r.type == Param::INT)
+        pi = new AutoBuffer<int64, 1>(*r.pi);
+    else if (r.type == Param::STRING)
+        ps = new AutoBuffer<String, 1>(*r.ps);
+    else if (r.type == Param::REAL)
+        pd = new AutoBuffer<double, 1>(*r.pd);
 }
 
 inline bool DictValue::isString() const
 {
-    return (type == cv::Param::STRING);
+    return (type == Param::STRING);
 }
 
 inline bool DictValue::isInt() const
 {
-    return (type == cv::Param::INT);
+    return (type == Param::INT);
+}
+
+inline bool DictValue::isReal() const
+{
+    return (type == Param::REAL || type == Param::INT);
+}
+
+inline int DictValue::size() const
+{
+    switch (type)
+    {
+    case Param::INT:
+        return (int)pi->size();
+        break;
+    case Param::STRING:
+        return (int)ps->size();
+        break;
+    case Param::REAL:
+        return (int)pd->size();
+        break;
+    default:
+        return -1;
+    }
 }
 
 }

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

@@ -7,62 +7,12 @@
 
 #include <opencv2/core.hpp>
 #include <opencv2/dnn/dict.hpp>
+#include <opencv2/dnn/blob.hpp>
 
 namespace cv
 {
 namespace dnn
 {
-    class Layer;
-    class NetConfiguration;
-    class Net;
-    class Blob;
-    class LayerParams;
-
-    //wrapper over cv::Mat and cv::UMat
-    class CV_EXPORTS Blob
-    {
-    public:
-        explicit Blob();
-        explicit Blob(InputArray in);
-
-        void create(int ndims, const int *sizes, int type = CV_32F);
-        void create(Vec4i shape, int type = CV_32F);
-        void create(int num, int cn, int rows, int cols, int type = CV_32F);
-
-        void fill(InputArray in);
-        void fill(int ndims, const int *sizes, int type, void *data, bool deepCopy = true);
-
-        Mat& getMatRef();
-        const Mat& getMatRef() const;
-        Mat getMat();
-        Mat getMat(int num, int channel);
-
-        //shape getters
-        int cols() const;
-        int rows() const;
-        int channels() const;
-        int num() const;
-        Size size2() const;
-        Vec4i shape() const;
-        int size(int index) const;
-        size_t total(int startAxis = 0, int endAxis = -1) const;
-
-        uchar *rawPtr(int num = 0, int cn = 0, int row = 0, int col = 0);
-
-        template<typename TFloat>
-        TFloat *ptr(int num = 0, int cn = 0, int row = 0, int col = 0);
-
-        int type() const;
-        bool isFloat() const;
-        bool isDouble() const;
-
-    private:
-        const int *sizes() const;
-        int dims() const;
-
-        Mat m;
-    };
-
     class CV_EXPORTS LayerParams : public Dict
     {
     public:
@@ -70,30 +20,11 @@ namespace dnn
         std::vector<Blob> learnedBlobs;
     };
 
-    class CV_EXPORTS LayerRegister
-    {
-    public:
-
-        typedef Ptr<Layer> (*Constuctor)(LayerParams &params);
-
-        static void registerLayer(const String &type, Constuctor constructor);
-
-        static void unregisterLayer(const String &type);
-
-        static Ptr<Layer> createLayerInstance(const String &type, LayerParams& params);
-
-    private:
-        LayerRegister();
-
-        struct Impl;
-        static Ptr<Impl> impl;
-    };
-
-    //this class allows to build new Layers
+    //Interface class allows to build new Layers
     class CV_EXPORTS Layer
     {
     public:
-        //TODO: this field must be declared as public if we want support possibility to change these params in runtime
+        //learned params of layer must be stored here to allow externally read them
         std::vector<Blob> learnedParams;
 
         virtual ~Layer();
@@ -103,16 +34,13 @@ namespace dnn
 
         virtual void forward(std::vector<Blob*> &inputs, std::vector<Blob> &outputs) = 0;
 
-        virtual int getNumInputs();
-        virtual int getNumOutputs();
         //each input/output can be labeled to easily identify their using "layer_name.output_name"
-        virtual String getInputName(int inputNum);
-        virtual String getOutputName(int outputNum);
+        virtual int inputNameToIndex(String inputName);
+        virtual int outputNameToIndex(String outputName);
     };
 
     //containers for String and int
     typedef DictValue LayerId;
-    typedef DictValue BlobId;
 
     class CV_EXPORTS Net
     {
@@ -125,14 +53,10 @@ namespace dnn
         int getLayerId(LayerId layer);
         void deleteLayer(LayerId layer);
 
-        //each output of each layer can be labeled by unique string label (as in Caffe)
-        //if label not specified then %layer_name%.%layer_output_id% can be used
-        void setOutputNames(LayerId layer, const std::vector<String> &outputNames);
-        void setLayerInputs(const std::vector<String> &outputs, LayerId layer);
         void setNetInputs(const std::vector<String> &inputBlobNames);
 
-        void connect(BlobId input, BlobId output);
-        void connect(const std::vector<BlobId> &outputs, const std::vector<BlobId> &inputs);
+        void connect(String outPin, String inpPin);
+        void connect(int outLayerId, int outNum, int inLayerId, int inNum);
 
         void forward();
         void forward(LayerId toLayer);
@@ -143,11 +67,11 @@ namespace dnn
         void forwardOpt(LayerId toLayer);
         void forwardOpt(const std::vector<LayerId> &toLayers);
 
-        void setBlob(BlobId outputName, const Blob &blob);
-        Blob getBlob(BlobId outputName);
+        void setBlob(String outputName, const Blob &blob);
+        Blob getBlob(String outputName);
 
         void setParam(LayerId layer, int numParam, const Blob &blob);
-        void getParam(LayerId layer, int numParam);
+        Blob getParam(LayerId layer, int numParam = 0);
 
     private:
 
@@ -164,8 +88,27 @@ namespace dnn
         virtual ~Importer();
     };
 
-    CV_EXPORTS Ptr<Importer> createCaffeImporter(const String &prototxt, const String &caffeModel);
+    CV_EXPORTS Ptr<Importer> createCaffeImporter(const String &prototxt, const String &caffeModel = String());
+
+    //Layer factory allows to create instances of registered layers.
+    class CV_EXPORTS LayerRegister
+    {
+    public:
+
+        typedef Ptr<Layer>(*Constuctor)(LayerParams &params);
 
+        static void registerLayer(const String &type, Constuctor constructor);
+
+        static void unregisterLayer(const String &type);
+
+        static Ptr<Layer> createLayerInstance(const String &type, LayerParams& params);
+
+    private:
+        LayerRegister();
+
+        struct Impl;
+        static Ptr<Impl> impl;
+    };
 
     //allows automatically register created layer on module load time
     struct _LayerRegisterer

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

@@ -1,5 +1,5 @@
-#ifndef __OPENCV_DNN_INL_HPP__
-#define __OPENCV_DNN_INL_HPP__
+#ifndef __OPENCV_DNN_DNN_INL_HPP__
+#define __OPENCV_DNN_DNN_INL_HPP__
 
 #include <opencv2/dnn.hpp>
 
@@ -7,123 +7,7 @@ namespace cv
 {
 namespace dnn
 {
-    inline Mat& Blob::getMatRef()
-    {
-        return m;
-    }
-
-    inline const Mat& Blob::getMatRef() const
-    {
-        return m;
-    }
-
-    inline Mat Blob::getMat()
-    {
-        return m;
-    }
-
-    inline Mat Blob::getMat(int num, int channel)
-    {
-        CV_Assert(0 <= num && num < this->num() && 0 <= channel && channel < this->channels());
-        return Mat(rows(), cols(), m.type(), this->rawPtr(num, channel));
-    }
-
-    inline int Blob::cols() const
-    {
-        CV_DbgAssert(m.dims > 2);
-        return m.size[m.dims-1];
-    }
-
-    inline int Blob::rows() const
-    {
-        CV_DbgAssert(m.dims > 2);
-        return m.size[m.dims-2];
-    }
-
-    inline Size Blob::size2() const
-    {
-        return Size(cols(), rows());
-    }
-
-    inline int Blob::channels() const
-    {
-        CV_DbgAssert(m.dims >= 3);
-        return m.size[m.dims-3];
-    }
-
-    inline int Blob::num() const
-    {
-        CV_DbgAssert(m.dims == 4);
-        return m.size[0];
-    }
-
-    inline Vec4i Blob::shape() const
-    {
-        CV_DbgAssert(m.dims == 4);
-        return Vec4i(m.size.p);
-    }
-
-    inline int Blob::size(int index) const
-    {
-        CV_Assert(index >= 0 && index < dims());
-        return sizes()[index];
-    }
-
-    inline size_t Blob::total(int startAxis, int endAxis) const
-    {
-        if (endAxis == -1)
-            endAxis = dims();
-
-        CV_Assert(0 <= startAxis && startAxis <= endAxis && endAxis <= dims());
-
-        size_t size = 1; //assume that blob isn't empty
-        for (int i = startAxis; i < endAxis; i++)
-            size *= (size_t) sizes()[i];
-
-        return size;
-    }
-
-    inline uchar* Blob::rawPtr(int num, int cn, int row, int col)
-    {
-        CV_DbgAssert(m.dims == 4);
-        return m.data + num * m.step[0] + cn * m.step[1] + row * m.step[2] + col * m.step[3];
-    }
-
-    template<typename TFloat>
-    TFloat *Blob::ptr(int n, int cn, int row, int col)
-    {
-        CV_Assert(m.type() == cv::DataType<TFloat>::type);
-        CV_Assert(0 <= n && n < num() && 0 <= cn && cn < channels() && 0 <= row && row < rows() && 0 <= col && col < cols());
-        return (TFloat*) rawPtr(n, cn, row, col);
-    }
-
-    inline int Blob::type() const
-    {
-        return m.depth();
-    }
-
-    inline bool Blob::isFloat() const
-    {
-        return (type() == CV_32F);
-    }
-
-    inline bool Blob::isDouble() const
-    {
-        return (type() == CV_32F);
-    }
-
-    inline const int * Blob::sizes() const
-    {
-        return &m.size[0];
-    }
-
-    inline int Blob::dims() const
-    {
-        return m.dims;
-    }
-
-
-
+    //code is absent ... today
 }
 }
 

modules/dnn/samples/classify_with_googlenet.cpp

@@ -25,50 +25,49 @@ std::vector<String> CLASES_NAMES;
 
 void initClassesNames()
 {
-    std::ifstream fp("ILSVRC2012_synsets.txt");
+    std::ifstream fp("synset_words.txt");
     CV_Assert(fp.is_open());
 
     std::string name;
     while (!fp.eof())
     {
         std::getline(fp, name);
-        CLASES_NAMES.push_back(name);
+        if (name.length())
+            CLASES_NAMES.push_back( name.substr(name.find(' ')+1) );
     }
 
-    CV_Assert(CLASES_NAMES.size() == 1000);
-
     fp.close();
 }
 
 
-int main(void)
+int main(int argc, char **argv)
 {
     Net net;
     {
-        Ptr<Importer> importer = createCaffeImporter("bvlc_alexnet.prototxt", "bvlc_alexnet.caffemodel");
+        Ptr<Importer> importer = createCaffeImporter("bvlc_googlenet.prototxt", "bvlc_googlenet.caffemodel");
         importer->populateNet(net);
     }
 
-    Mat img = imread("zebra.jpg");
+    String filename = (argc > 1) ? argv[1] : "space_shuttle.jpg";
+
+    Mat img = imread(filename);
     CV_Assert(!img.empty());
     cvtColor(img, img, COLOR_BGR2RGB);
-    img.convertTo(img, CV_32F);
     resize(img, img, Size(227, 227));
-    subtract(img, cv::mean(img), img);
     Blob imgBlob(img);
 
-    net.setBlob("data", imgBlob);
+    net.setBlob(".data", imgBlob);
     net.forward();
 
-    Blob probBlob = net.getBlob("prob");
-    ClassProb bc = getMaxClass(probBlob);
+    Blob prob = net.getBlob("prob");
+    ClassProb bc = getMaxClass(prob);
 
     initClassesNames();
     std::string className = (bc.first < (int)CLASES_NAMES.size()) ? CLASES_NAMES[bc.first] : "unnamed";
 
     std::cout << "Best class:";
     std::cout << " #" << bc.first;
-    std::cout << " (from " << probBlob.total(1) << ")";
+    std::cout << " (from " << prob.total(1) << ")";
     std::cout << " \"" + className << "\"";
     std::cout <<  std::endl;
     std::cout << "Prob: " << bc.second * 100 << "%" << std::endl;

modules/dnn/scripts/download_model.py

@@ -0,0 +1,79 @@
+#!/usr/bin/env python
+import os
+import sys
+import time
+import urllib
+import hashlib
+import argparse
+import json
+
+
+def reporthook(count, block_size, total_size):
+    """
+    From http://blog.moleculea.com/2012/10/04/urlretrieve-progres-indicator/
+    """
+    global start_time
+    global prev_duration
+    if count == 0:
+        start_time = time.time()
+        prev_duration = -1
+        return
+    duration = max(1, time.time() - start_time)
+    if int(duration) == int(prev_duration):
+        return
+
+    progress_size = int(count * block_size)
+    speed = int(progress_size / (1024 * duration))
+    percent = int(count * block_size * 100 / total_size)
+    sys.stdout.write("\r...%d%%, %d MB, %d KB/s, %d seconds passed" %
+                     (percent, progress_size / (1024 * 1024), speed, duration))
+    sys.stdout.flush()
+    prev_duration = duration
+
+
+# Function for checking SHA1.
+def model_checks_out(filename, sha1):
+    with open(filename, 'r') as f:
+        return hashlib.sha1(f.read()).hexdigest() == sha1
+
+def model_download(filename, url, sha1):
+    # Check if model exists.
+    if os.path.exists(filename) and model_checks_out(filename, sha1):
+        print("Model {} already exists.".format(filename))
+        return
+
+    # Download and verify model.
+    urllib.urlretrieve(url, filename, reporthook)
+    print model_checks_out(filename, sha1)
+    if not model_checks_out(filename, sha1):
+        print("ERROR: model {} did not download correctly!".format(url))
+        sys.exit(1)
+
+if __name__ == '__main__':
+    parser = argparse.ArgumentParser(description="Downloading trained model binaries.")
+    parser.add_argument("download_list")
+    args = parser.parse_args()
+
+    test_dir = os.environ.get("OPENCV_TEST_DATA_PATH")
+    if not test_dir:
+        print "ERROR: OPENCV_TEST_DATA_PATH environment not specified"
+        sys.exit(1)
+
+    try:
+        with open(args.download_list, 'r') as f:
+            models_to_download = json.load(f)
+    except:
+        print "ERROR: Can't pasrse {}".format(args.download_list)
+        sys.exit(1)
+
+    for model_name in models_to_download:
+        model = models_to_download[model_name]
+
+        dst_dir = os.path.join(test_dir, os.path.dirname(model['file']))
+        dst_file = os.path.join(test_dir, model['file'])
+        if not os.path.exists(dst_dir):
+            print "ERROR: Can't find module testdata path '{}'".format(dst_dir)
+            sys.exit(1)
+
+        print "Downloading model '{}' to {} from {} ...".format(model_name, dst_file, model['url'])
+        model_download(dst_file, model['url'], model['sha1'])

modules/dnn/scripts/test_models.json

@@ -0,0 +1,7 @@
+{
+  "googlenet": {
+    "file": "dnn/bvlc_googlenet.caffemodel",
+    "url": "http://dl.caffe.berkeleyvision.org/bvlc_googlenet.caffemodel",
+    "sha1": "405fc5acd08a3bb12de8ee5e23a96bec22f08204"
+  }
+}

modules/dnn/src/blob.cpp

@@ -0,0 +1,180 @@
+#include "precomp.hpp"
+
+namespace cv
+{
+namespace dnn
+{
+
+    Blob::Blob()
+    {
+        int zeros[4] = { 0, 0, 0, 0 };
+        m = Mat(4, zeros, CV_32F, NULL);
+    }
+
+    static inline int getMatChannels(const Mat &mat)
+    {
+       return (mat.dims <= 2) ? mat.channels() : mat.size[0];
+    }
+
+    static BlobShape getBlobShpae(std::vector<Mat> &vmat, int requestedCn = -1)
+    {
+        BlobShape shape(4);
+        int cnSum = 0, matCn;
+
+        CV_Assert(vmat.size() > 0);
+
+        for (size_t i = 0; i < vmat.size(); i++)
+        {
+            Mat &mat = vmat[i];
+            CV_Assert(!mat.empty());
+            CV_Assert((mat.dims == 3 && mat.channels() == 1) || mat.dims <= 2);
+
+            matCn = getMatChannels(mat);
+            cnSum += getMatChannels(mat);
+
+            if (i == 0)
+            {
+                shape[-1] = mat.cols;
+                shape[-2] = mat.rows;
+                shape[-3] = (requestedCn <= 0) ? matCn : requestedCn;
+            }
+            else
+            {
+                if (mat.cols != shape[-1] || mat.rows != shape[-2])
+                    CV_Error(Error::StsError, "Each Mat.size() must be equal");
+
+                if (requestedCn <= 0 && matCn != shape[-3])
+                    CV_Error(Error::StsError, "Each Mat.chnannels() (or number of planes) must be equal");
+            }
+        }
+
+        if (cnSum % shape[-3] != 0)
+            CV_Error(Error::StsError, "Total number of channels in vector is not a multiple of requsted channel number");
+
+        shape[0] = cnSum / shape[-3];
+        return shape;
+    }
+
+    static std::vector<Mat> extractMatVector(InputArray in)
+    {
+        if (in.isMat() || in.isUMat())
+        {
+            return std::vector<Mat>(1, in.getMat());
+        }
+        else if (in.isMatVector())
+        {
+            return *static_cast<const std::vector<Mat>*>(in.getObj());
+        }
+        else if (in.isUMatVector())
+        {
+            std::vector<Mat> vmat;
+            in.getMatVector(vmat);
+            return vmat;
+        }
+        else
+        {
+            CV_Assert(in.isMat() || in.isMatVector() || in.isUMat() || in.isUMatVector());
+            return std::vector<Mat>();
+        }
+    }
+
+    Blob::Blob(InputArray in, int dstCn)
+    {
+        CV_Assert(dstCn == -1 || dstCn > 0);
+        std::vector<Mat> inMats = extractMatVector(in);
+        BlobShape dstShape = getBlobShpae(inMats, dstCn);
+
+        m.create(dstShape.dims(), dstShape.ptr(), CV_32F);
+
+        std::vector<Mat> wrapBuf(dstShape[-3]);
+        int elemSize = (int)m.elemSize();
+        uchar *ptr = this->ptrRaw();
+        for (size_t i = 0; i < inMats.size(); i++)
+        {
+            Mat inMat = inMats[i];
+
+            if (inMat.dims <= 2)
+            {
+                inMat.convertTo(inMat, m.type());
+
+                wrapBuf.resize(0);
+                for (int cn = 0; cn < inMat.channels(); cn++)
+                {
+                    wrapBuf.push_back(Mat(inMat.rows, inMat.cols, m.type(), ptr));
+                    ptr += elemSize * inMat.total();
+                }
+
+                cv::split(inMat, wrapBuf);
+            }
+            else
+            {
+                inMat.convertTo(Mat(inMat.dims, inMat.size, m.type(), ptr), m.type());
+                ptr += elemSize * inMat.total();
+            }
+        }
+    }
+
+    void Blob::fill(const BlobShape &shape, int type, void *data, bool deepCopy)
+    {
+        CV_Assert(type == CV_32F || type == CV_64F);
+
+        if (deepCopy)
+        {
+            m.create(shape.dims(), shape.ptr(), type);
+            memcpy(m.data, data, m.total() * m.elemSize());
+        }
+        else
+        {
+            m = Mat(shape.dims(), shape.ptr(), type, data);
+        }
+    }
+
+    void Blob::fill(InputArray in)
+    {
+        CV_Assert(in.isMat() || in.isMatVector());
+
+        //TODO
+        *this = Blob(in);
+    }
+
+    void Blob::create(const BlobShape &shape, int type)
+    {
+        CV_Assert(type == CV_32F || type == CV_64F);
+        m.create(shape.dims(), shape.ptr(), type);
+    }
+
+    inline void squeezeShape(const int srcDims, const int *srcSizes, const int dstDims, int *dstSizes)
+    {
+        const int m = std::min(dstDims, srcDims);
+
+        //copy common(last) dimensions
+        for (int i = 0; i < m; i++)
+            dstSizes[dstDims - 1 - i] = srcSizes[srcDims - 1 - i];
+
+        //either flatten extra dimensions
+        for (int i = m; i < srcDims; i++)
+            dstSizes[0] *= srcSizes[srcDims - 1 - i];
+
+        //either fill gaps
+        for (int i = m; i < dstDims; i++)
+            dstSizes[dstDims - 1 - i] = 1;
+    }
+
+    Vec4i Blob::shape4() const
+    {
+        return Vec4i(num(), channels(), rows(), cols());
+    }
+
+    std::ostream &operator<< (std::ostream &stream, const BlobShape &shape)
+    {
+        stream << "[";
+
+        for (int i = 0; i < shape.dims() - 1; i++)
+            stream << shape[i] << ", ";
+        if (shape.dims() > 0)
+            stream << shape[-1];
+
+        return stream << "]";
+    }
+}
+}

modules/dnn/src/caffe/glog_emulator.hpp

@@ -1,4 +1,5 @@
-#pragma once
+#ifndef __OPENCV_DNN_CAFFE_GLOG_EMULATOR__
+#define __OPENCV_DNN_CAFFE_GLOG_EMULATOR__
 #include <stdlib.h>
 #include <iostream>
 #include <sstream>
@@ -52,3 +53,4 @@ public:
 };
 
 }
+#endif

modules/dnn/src/caffe_importer.cpp

@@ -37,70 +37,79 @@ namespace
                 ReadNetParamsFromBinaryFileOrDie(caffeModel, &netBinary);
         }
 
-        inline bool skipCaffeLayerParam(const FieldDescriptor *fd)
-        {
-            const std::string &name = fd->name();
-
-            if (fd->cpp_type() != FieldDescriptor::CPPTYPE_MESSAGE)
-            {
-                static const char *SKIP_FIELDS[] = { "type", "name", "top", "bottom", NULL };
-
-                for (int i = 0; SKIP_FIELDS[i]; i++)
-                {
-                    if (name == SKIP_FIELDS[i])
-                        return true;
-                }
-
-                return false;
-            }
-            else
-            {
-                static const std::string _param("_param");
-                bool endsWith_param = (name.size() >= _param.size()) && name.compare(name.size() - _param.size(), _param.size(), _param) == 0;
-                return !endsWith_param;
-            }
-        }
-
         void addParam(const Message &msg, const FieldDescriptor *field, cv::dnn::LayerParams &params)
         {
-            const Reflection *msgRefl = msg.GetReflection();
+            const Reflection *refl = msg.GetReflection();
             int type = field->cpp_type();
             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))
+            #define SET_UP_FILED(getter, arrayConstr, gtype)                                    \
+                if (isRepeated) {                                                               \
+                    const RepeatedField<gtype> &v = refl->GetRepeatedField<gtype>(msg, field);  \
+                    params.set(name, DictValue::arrayConstr(v.begin(), (int)v.size()));                  \
+                }                                                                               \
+                else {                                                                          \
+                    params.set(name, refl->getter(msg, field));                               \
+                }
 
             switch (type)
             {
             case FieldDescriptor::CPPTYPE_INT32:
-                std::cout << params.set(name, GET_FIRST(Int32));
+                SET_UP_FILED(GetInt32, arrayInt, ::google::protobuf::int32);
                 break;
             case FieldDescriptor::CPPTYPE_UINT32:
-                std::cout << params.set(name, GET_FIRST(UInt32));
+                SET_UP_FILED(GetUInt32, arrayInt, ::google::protobuf::uint32);
+                break;
+            case FieldDescriptor::CPPTYPE_INT64:
+                SET_UP_FILED(GetInt32, arrayInt, ::google::protobuf::int64);
+                break;
+            case FieldDescriptor::CPPTYPE_UINT64:
+                SET_UP_FILED(GetUInt32, arrayInt, ::google::protobuf::uint64);
+                break;
+            case FieldDescriptor::CPPTYPE_BOOL:
+                SET_UP_FILED(GetBool, arrayInt, bool);
                 break;
             case FieldDescriptor::CPPTYPE_DOUBLE:
-                std::cout << params.set(name, GET_FIRST(Double));
+                SET_UP_FILED(GetDouble, arrayReal, double);
                 break;
             case FieldDescriptor::CPPTYPE_FLOAT:
-                std::cout << params.set(name, GET_FIRST(Float));
+                SET_UP_FILED(GetFloat, arrayReal, float);
                 break;
-            case FieldDescriptor::CPPTYPE_ENUM:
-                std::cout << params.set(name, GET_FIRST(Enum)->name());
+            case FieldDescriptor::CPPTYPE_STRING:
+                if (isRepeated) {
+                    const RepeatedPtrField<std::string> &v = refl->GetRepeatedPtrField<std::string>(msg, field);
+                    params.set(name, DictValue::arrayString(v.begin(), (int)v.size()));
+                }
+                else {
+                    params.set(name, refl->GetString(msg, field));
+                }
                 break;
-            case FieldDescriptor::CPPTYPE_BOOL:
-                std::cout << params.set(name, GET_FIRST(Bool));
+            case FieldDescriptor::CPPTYPE_ENUM:
+                if (isRepeated) {
+                    int size = refl->FieldSize(msg, field);
+                    std::vector<cv::String> buf(size);
+                    for (int i = 0; i < size; i++)
+                        buf[i] = refl->GetRepeatedEnum(msg, field, i)->name();
+                    params.set(name, DictValue::arrayString(buf.begin(), size));
+                }
+                else {
+                    params.set(name, refl->GetEnum(msg, field)->name());
+                }
                 break;
             default:
-                std::cout << "unknown";
+                CV_Error(Error::StsError, "Unknown type \"" + String(field->type_name()) + "\" in prototxt");
                 break;
             }
+        }
 
-            std::cout << std::endl;
+        inline static bool ends_with_param(const std::string &str)
+        {
+            static const std::string _param("_param");
+            return (str.size() >= _param.size()) && str.compare(str.size() - _param.size(), _param.size(), _param) == 0;
         }
 
-        void extractLayerParams(const Message &msg, cv::dnn::LayerParams &params)
+        void extractLayerParams(const Message &msg, cv::dnn::LayerParams &params, bool isInternal = false)
         {
             const Descriptor *msgDesc = msg.GetDescriptor();
             const Reflection *msgRefl = msg.GetReflection();
@@ -109,19 +118,21 @@ namespace
             {
                 const FieldDescriptor *fd = msgDesc->field(fieldId);
 
+                if (!isInternal && !ends_with_param(fd->name()))
+                    continue;
+
                 bool hasData =  fd->is_required() ||
-                               (fd->is_optional() && (msgRefl->HasField(msg, fd) /*|| fd->has_default_value()*/)) ||
+                               (fd->is_optional() && msgRefl->HasField(msg, fd)) ||
                                (fd->is_repeated() && msgRefl->FieldSize(msg, fd) > 0);
-
-                if ( !hasData || skipCaffeLayerParam(fd) )
+                if (!hasData)
                     continue;
 
                 if (fd->cpp_type() == FieldDescriptor::CPPTYPE_MESSAGE)
                 {
                     if (fd->is_repeated()) //Extract only first item!
-                        extractLayerParams(msgRefl->GetRepeatedMessage(msg, fd, 0), params);
+                        extractLayerParams(msgRefl->GetRepeatedMessage(msg, fd, 0), params, true);
                     else
-                        extractLayerParams(msgRefl->GetMessage(msg, fd), params);
+                        extractLayerParams(msgRefl->GetMessage(msg, fd), params, true);
                 }
                 else
                 {
@@ -130,39 +141,41 @@ namespace
             }
         }
 
-        void blobFromProto(const caffe::BlobProto &protoBlob, cv::dnn::Blob &dstBlob)
+        BlobShape blobShapeFromProto(const caffe::BlobProto &pbBlob)
         {
-            AutoBuffer<int, 4> shape;
-
-            if (protoBlob.has_num() || protoBlob.has_channels() || protoBlob.has_height() || protoBlob.has_width())
+            if (pbBlob.has_num() || pbBlob.has_channels() || pbBlob.has_height() || pbBlob.has_width())
             {
-                shape.resize(4);
-                shape[0] = protoBlob.num();
-                shape[1] = protoBlob.channels();
-                shape[2] = protoBlob.height();
-                shape[3] = protoBlob.width();
+                return BlobShape(pbBlob.num(), pbBlob.channels(), pbBlob.height(), pbBlob.width());
             }
-            else if (protoBlob.has_shape())
+            else if (pbBlob.has_shape())
             {
-                const caffe::BlobShape &_shape = protoBlob.shape();
-                shape.resize(_shape.dim_size());
+                const caffe::BlobShape &_shape = pbBlob.shape();
+                BlobShape shape(_shape.dim_size());
 
                 for (int i = 0; i < _shape.dim_size(); i++)
-                    shape[i] = _shape.dim(i);
+                    shape[i] = (int)_shape.dim(i);
+
+                return shape;
             }
             else
             {
-                CV_Error(cv::Error::StsAssert, "Unknown shape of input blob");
+                CV_Error(Error::StsError, "Unknown shape of input blob");
+                return BlobShape(-1);
+            }
         }
 
-            dstBlob.create(shape.size(), shape, CV_32F);
-            CV_Assert(protoBlob.data_size() == (int)dstBlob.getMatRef().total());
+        void blobFromProto(const caffe::BlobProto &pbBlob, cv::dnn::Blob &dstBlob)
+        {
+            BlobShape shape = blobShapeFromProto(pbBlob);
+
+            dstBlob.create(shape, CV_32F);
+            CV_Assert(pbBlob.data_size() == (int)dstBlob.getMatRef().total());
 
-            CV_DbgAssert(protoBlob.GetDescriptor()->FindFieldByLowercaseName("data")->cpp_type() == FieldDescriptor::CPPTYPE_FLOAT);
+            CV_DbgAssert(pbBlob.GetDescriptor()->FindFieldByLowercaseName("data")->cpp_type() == FieldDescriptor::CPPTYPE_FLOAT);
             float *dstData = dstBlob.getMatRef().ptr<float>();
 
-            for (int i = 0; i < protoBlob.data_size(); i++)
-                dstData[i] = protoBlob.data(i);
+            for (int i = 0; i < pbBlob.data_size(); i++)
+                dstData[i] = pbBlob.data(i);
         }
 
         void extractBinaryLayerParms(const caffe::LayerParameter& layer, LayerParams& layerParams)
@@ -187,55 +200,100 @@ namespace
             }
         }
 
+        struct BlobNote
+        {
+            BlobNote(const std::string &_name, int _layerId, int _outNum) :
+                name(_name.c_str()), layerId(_layerId), outNum(_outNum) {}
+
+            const char *name;
+            int layerId, outNum;
+        };
+
         void populateNet(Net dstNet)
         {
+            int layersSize = net.layer_size();
+            std::vector<BlobNote> addedBlobs;
+            addedBlobs.reserve(layersSize + 1);
+
             //setup input layer names
             {
                 std::vector<String> netInputs(net.input_size());
-                for (int ii = 0; ii < net.input_size(); ii++)
-                    netInputs[ii] = net.input(ii);
+                for (int inNum = 0; inNum < net.input_size(); inNum++)
+                {
+                    addedBlobs.push_back(BlobNote(net.input(inNum), 0, inNum));
+                    netInputs[inNum] = net.input(inNum);
+                }
                 dstNet.setNetInputs(netInputs);
             }
 
-            int layersSize = net.layer_size();
-
-            std::vector<String> layersName(layersSize);
-            std::vector<int> layersId(layersSize);
-            std::vector<std::vector<String> > bottomsVec(layersSize);
-
             for (int li = 0; li < layersSize; li++)
             {
-                const caffe::LayerParameter layer = net.layer(li);
+                const caffe::LayerParameter &layer = net.layer(li);
                 String name = layer.name();
                 String type = layer.type();
                 LayerParams layerParams;
 
-                std::vector<String> tops;
-                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);
 
                 int id = dstNet.addLayer(name, type, layerParams);
-                dstNet.setOutputNames(id, tops);
 
-                layersName[li] = name;
-                layersId[li] = id;
+                for (int inNum = 0; inNum < layer.bottom_size(); inNum++)
+                    addInput(layer.bottom(inNum), id, inNum, dstNet, addedBlobs);
+
+                for (int outNum = 0; outNum < layer.top_size(); outNum++)
+                    addOutput(layer, id, outNum, addedBlobs);
+            }
         }
 
-            for (int li = 0; li < layersSize; li++)
+        void addOutput(const caffe::LayerParameter &layer, int layerId, int outNum, std::vector<BlobNote> &addedBlobs)
+        {
+            const std::string &name = layer.top(outNum);
+
+            bool haveDups = false;
+            for (int idx = (int)addedBlobs.size() - 1; idx >= 0; idx--)
+            {
+                if (addedBlobs[idx].name == name)
                 {
-                dstNet.setLayerInputs(bottomsVec[li], layersId[li]);
+                    haveDups = true;
+                    break;
+                }
+            }
+
+            if (haveDups)
+            {
+                bool isInplace = layer.bottom_size() > outNum && layer.bottom(outNum) == name;
+                if (!isInplace)
+                    CV_Error(Error::StsBadArg, "Duplicate blobs produced by multiple sources");
             }
+
+            addedBlobs.push_back(BlobNote(name, layerId, outNum));
+        }
+
+        void addInput(const std::string &name, int layerId, int inNum, Net &dstNet, std::vector<BlobNote> &addedBlobs)
+        {
+            int idx;
+            for (idx = (int)addedBlobs.size() - 1; idx >= 0; idx--)
+            {
+                if (addedBlobs[idx].name == name)
+                    break;
+            }
+
+            if (idx < 0)
+            {
+                CV_Error(Error::StsObjectNotFound, "Can't found output blob \"" + name + "\"");
+                return;
+            }
+
+            dstNet.connect(addedBlobs[idx].layerId, addedBlobs[idx].outNum, layerId, inNum);
         }
 
         ~CaffeImporter()
         {
 
         }
+
+
     };
 
 }

modules/dnn/src/dnn.cpp

@@ -2,6 +2,7 @@
 #include <set>
 #include <algorithm>
 #include <iostream>
+#include <sstream>
 
 using namespace cv;
 using namespace cv::dnn;
@@ -16,172 +17,111 @@ namespace cv
 namespace dnn
 {
 
-Blob::Blob()
+template<typename T>
+String toString(const T &v)
 {
-    int zeros[4] = {0, 0, 0, 0};
-    m = Mat(4, zeros, CV_32F, NULL);
+    std::ostringstream ss;
+    ss << v;
+    return ss.str();
 }
 
-Blob::Blob(InputArray in)
+struct LayerPin
 {
-    CV_Assert(in.isMat() || in.isUMat());
-
-    if (in.isMat())
-    {
-        Mat mat = in.getMat();
-
-        CV_Assert(mat.dims == 2);
-        int rows = mat.rows;
-        int cols = mat.cols;
-        int cn = mat.channels();
-        int type = mat.type();
-        int dstType = CV_MAKE_TYPE(CV_MAT_DEPTH(type), 1);
+    int lid;
+    int oid;
 
-        int size[3] = { cn, rows, cols };
-        this->create(3, size, dstType);
-        uchar *data = m.data;
-        int step = rows * cols * CV_ELEM_SIZE(dstType);
+    LayerPin(int layerId = -1, int outputId = -1)
+        : lid(layerId), oid(outputId) {}
 
-        if (cn == 1)
+    bool valid() const
     {
-            Mat wrapper2D(rows, cols, dstType, m.data);
-            mat.copyTo(wrapper2D);
-        }
-        else
-        {
-            std::vector<Mat> wrappers(cn);
-            for (int i = 0; i < cn; i++)
-            {
-                wrappers[i] = Mat(rows, cols, dstType, data);
-                data += step;
+        return (lid >= 0 && oid >= 0);
     }
 
-            cv::split(mat, wrappers);
-        }
-    }
-    else
+    bool equal(const LayerPin &r) const
     {
-        CV_Error(cv::Error::StsNotImplemented, "Not Implemented");
+        return (lid == r.lid && oid == r.oid);
     }
-}
+};
 
-static Vec4i blobNormalizeShape(int ndims, const int *sizes)
+struct LayerData
 {
-    Vec4i shape = Vec4i::all(1);
-
-    for (int i = 0; i < std::min(3, ndims); i++)
-        shape[3 - i] = sizes[ndims-1 - i];
+    LayerData() {}
+    LayerData(int _id, const String &_name, const String &_type, LayerParams &_params)
+        : id(_id), name(_name), type(_type), params(_params) {}
 
-    for (int i = 3; i < ndims; i++)
-        shape[0] *= sizes[ndims-1 - i];
+    int id;
+    String name;
+    String type;
+    LayerParams params;
 
-    return shape;
-}
+    std::vector<LayerPin> inputBlobsId;
+    std::set<int> inputLayersId;
+    std::set<int> requiredOutputs;
 
-void Blob::fill(int ndims, const int *sizes, int type, void *data, bool deepCopy)
-{
-    CV_Assert(type == CV_32F || type == CV_64F);
+    Ptr<Layer> layerInstance;
+    std::vector<Blob> outputBlobs;
+    std::vector<Blob*> inputBlobs;
 
-    Vec4i shape = blobNormalizeShape(ndims, sizes);
+    int flag;
 
-    if (deepCopy)
+    Ptr<Layer> getLayerInstance()
     {
-        m.create(3, &shape[0], type);
-        size_t dataSize = m.total() * m.elemSize();
-        memcpy(m.data, data, dataSize);
-    }
-    else
+        if (layerInstance)
+            return layerInstance;
+
+        layerInstance = LayerRegister::createLayerInstance(type, params);
+        if (!layerInstance)
         {
-        m = Mat(shape.channels, &shape[0], type, data);
+            CV_Error(Error::StsError, "Can't create layer \"" + name + "\" of type \"" + type + "\"");
         }
-}
-
-void Blob::fill(InputArray in)
-{
-    CV_Assert(in.isMat() || in.isMatVector());
-
-    //TODO
-    *this = Blob(in);
-}
 
-void Blob::create(int ndims, const int *sizes, int type)
-{
-    CV_Assert(type == CV_32F || type == CV_64F);
-    Vec4i shape = blobNormalizeShape(ndims, sizes);
-    m.create(shape.channels, &shape[0], type);
-}
-
-void Blob::create(Vec4i shape, int type)
-{
-    m.create(shape.channels, &shape[0], type);
-}
-
-void Blob::create(int num, int cn, int rows, int cols, int type)
-{
-    Vec4i shape(num, cn, rows, cols);
-    create(4, &shape[0], type);
-}
-
-//////////////////////////////////////////////////////////////////////////
-
-struct LayerOutId
-{
-    int lid;
-    int oid;
-    String name;
-
-    LayerOutId() {}
-    LayerOutId(int layerId, int outputId, const String &outputName = String())
-        : lid(layerId), oid(outputId), name(outputName) {}
+        return layerInstance;
+    }
 };
 
-struct LayerData
+//fake layer containing network input blobs
+struct NetInputLayer : public Layer
 {
-    LayerData() {}
-    LayerData(const String &_name, const String &_type, LayerParams &_params)
-        : name(_name), type(_type), params(_params) {}
-
-    String name;
-    String type;
-    LayerParams params;
+    void allocate(const std::vector<Blob*>&, std::vector<Blob>&) {}
+    void forward(std::vector<Blob*>&, std::vector<Blob>&) {}
 
-    std::vector<String> outputNames;
-    std::vector<String> inputNames;
-    bool hasNamedOutput(const String &name)
+    int outputNameToIndex(String tgtName)
     {
-        return std::find(outputNames.begin(), outputNames.end(), name) != outputNames.end();
+        int idx = (int)(std::find(outNames.begin(), outNames.end(), tgtName) - outNames.begin());
+        return (idx < (int)outNames.size()) ? idx : -1;
     }
-    bool hasNemedInput(const String &name)
+
+    void setNames(const std::vector<String> &names)
     {
-        return std::find(inputNames.begin(), inputNames.end(), name) != inputNames.end();
+        outNames.assign(names.begin(), names.end());
     }
 
-    std::vector<LayerOutId> inputBlobsId;
-    std::set<int> inputLayersId;
-    std::set<int> requiredOutputs;
-
-    Ptr<Layer> layerInstance;
-    std::vector<Blob> outputBlobs;
-    std::vector<Blob*> inputBlobs;
-
-    int flag;
+private:
+    std::vector<String> outNames;
 };
 
 struct Net::Impl
 {
     Impl()
     {
-        LayerParams paramsEmpty;
-        layers.insert(make_pair(0, LayerData("_input", "_noType", paramsEmpty)));
+        //allocate fake net input layer
+        netInputLayer = Ptr<NetInputLayer>(new NetInputLayer());
+        LayerData &inpl = layers.insert( make_pair(0, LayerData()) ).first->second;
+        inpl.id = 0;
+        inpl.name = "_input";
+        inpl.type = "__NetInputLayer__";
+        inpl.layerInstance = netInputLayer;
+
         lastLayerId = 1;
         netWasAllocated = false;
     }
 
+    Ptr<NetInputLayer> netInputLayer;
     std::vector<int> netOutputs;
 
     typedef std::map<int, LayerData> MapIdToLayerData;
     std::map<int, LayerData> layers;
-
     std::map<String, int> layerNameToId;
 
     int lastLayerId;
@@ -192,9 +132,8 @@ struct Net::Impl
     {
         if (!netWasAllocated)
         {
-            connectInputs();
             allocateLayers();
-            computeNetOutputs();
+            computeNetOutputLayers();
 
             netWasAllocated = true;
         }
@@ -206,121 +145,130 @@ struct Net::Impl
         return (it != layerNameToId.end()) ? it->second : -1;
     }
 
-    int getLayerId(const DictValue &v)
-    {
-        if (v.isString())
-            return getLayerId(v.get<String>());
-        else if (v.isInt())
-            return v.get<int>();
-        else
+    int getLayerId(int id)
     {
-            CV_Assert(v.isString() || v.isInt());
-            return -1;
-        }
+        MapIdToLayerData::iterator it = layers.find(id);
+        return (it != layers.end()) ? id : -1;
     }
 
-    LayerData& getLayerData(const DictValue &v)
+    int getLayerId(DictValue &layerDesc)
     {
-        int id = getLayerId(v);
-        std::map<int, LayerData>::iterator it = layers.find(id);
-        CV_Assert(id >= 0 && it != layers.end());
-        return it->second;
+        if (layerDesc.isInt())
+            return getLayerId(layerDesc.get<int>());
+        else if (layerDesc.isString())
+            return getLayerId(layerDesc.get<String>());
+
+        CV_Assert(layerDesc.isInt() || layerDesc.isString());
+        return -1;
     }
 
-    int findOutputsByName(const String &name, LayerOutId *found, int maxCount = 1)
+    String getLayerName(int id)
     {
-        int count = 0;
+        MapIdToLayerData::iterator it = layers.find(id);
+        return (it != layers.end()) ? it->second.name : "(unknown layer)";
+    }
 
-        MapIdToLayerData::iterator it;
-        for (it = layers.begin(); it != layers.end() && count < maxCount; it++)
+    LayerData& getLayerData(int id)
     {
-            int lid = it->first;
-            LayerData &ld = it->second;
+        MapIdToLayerData::iterator it = layers.find(id);
 
-            for (size_t oi = 0; oi < ld.outputNames.size() && count < maxCount; oi++)
-            {
-                if (ld.outputNames[oi] == name)
-                    found[count++] = LayerOutId(lid, (int)oi);
-            }
-        }
+        if (it == layers.end())
+            CV_Error(Error::StsError, "Layer with requested id=" + toString(id) + " not found");
 
-        return count;
+        return it->second;
     }
 
-    void connectInputs()
+    LayerData& getLayerData(const String &layerName)
     {
-        LayerOutId foundOutputs[3], out;
+        int id = getLayerId(layerName);
 
-        MapIdToLayerData::iterator it;
-        for (it = layers.begin(); it != layers.end(); it++)
-        {
-            LayerData &ld = it->second;
+        if (id < 0)
+            CV_Error(Error::StsError, "Requsted layer \"" + layerName + "\" not found");
 
-            ld.inputBlobs.resize(ld.inputNames.size());
-            ld.inputBlobsId.resize(ld.inputNames.size());
-            ld.inputLayersId.clear();
+        return getLayerData(id);
+    }
 
-            for (size_t ii = 0; ii < ld.inputNames.size(); ii++)
+    LayerData& getLayerData(const DictValue &layerDesc)
     {
-                const String &tgtName = ld.inputNames[ii];
-
-                int foundCount = findOutputsByName(tgtName, foundOutputs, 3);
+        if (layerDesc.isInt())
+            return getLayerData(layerDesc.get<int>());
+        else if (layerDesc.isString())
+            return getLayerData(layerDesc.get<String>());
 
-                if (foundCount > 2)
-                {
-                    CV_Error(cv::Error::StsNotImplemented, "Two or more non-inplace blobs have the same name \"" + tgtName + "\"");
+        CV_Assert(layerDesc.isInt() || layerDesc.isString());
+        return *((LayerData*)NULL);
     }
-                else if (foundCount == 2)
-                {
-                    bool inPlace[2];
-                    inPlace[0] = layers[ foundOutputs[0].lid ].hasNemedInput(tgtName);
-                    inPlace[1] = layers[ foundOutputs[1].lid ].hasNemedInput(tgtName);
 
-                    if (!inPlace[0] && !inPlace[1])
+    static void addLayerInput(LayerData &ld, int inNum, LayerPin from)
     {
-                        CV_Error(cv::Error::StsNotImplemented, "Two or more non-inplace blobs have the same name \"" + tgtName + "\"");
-                    }
-                    else if (inPlace[0] && inPlace[1])
+        if ((int)ld.inputBlobsId.size() <= inNum)
         {
-                        CV_Error(cv::Error::StsNotImplemented, "Two or more blobs has same in-place blob \"" + tgtName + "\"");
+            ld.inputBlobsId.resize(inNum + 1);
         }
         else
         {
-                        if (ld.hasNamedOutput(tgtName))
-                            out = (inPlace[0]) ? foundOutputs[1] : foundOutputs[0];
-                        else
-                            out = (inPlace[0]) ? foundOutputs[0] : foundOutputs[1];
+            LayerPin storedFrom = ld.inputBlobsId[inNum];
+            if (storedFrom.valid() && !storedFrom.equal(from))
+                CV_Error(Error::StsError, "Input #" + toString(inNum) + "of layer \"" + ld.name + "\" already was connected");
         }
+
+        ld.inputBlobsId[inNum] = from;
     }
-                else if (foundCount == 0)
+
+    static void splitPin(const String &pinAlias, String &layerName, String &outName)
     {
-                    CV_Error(cv::Error::StsBadArg, "Can't find specified input blob \"" + tgtName + "\" for layer \"" + ld.name + "\"");
-                    continue;
+        size_t delimPos = pinAlias.find('.');
+        layerName = pinAlias.substr(0, delimPos);
+        outName = (delimPos == String::npos) ? String() : pinAlias.substr(delimPos + 1);
     }
-                else
+
+    int resolvePinOutputName(LayerData &ld, const String &outName, bool isOutPin)
     {
-                    out = foundOutputs[0];
-                }
+        if (outName.empty())
+            return 0;
+
+        if (std::isdigit(outName[0]))
+        {
+            char *lastChar;
+            long inum = std::strtol(outName.c_str(), &lastChar, 10);
 
-                ld.inputBlobsId[ii] = out;
-                ld.inputLayersId.insert(out.lid);
-                layers[out.lid].requiredOutputs.insert(out.oid);
+            if (*lastChar == 0)
+            {
+                CV_Assert(inum == (int)inum);
+                return (int)inum;
             }
         }
 
-        for (it = layers.begin(); it != layers.end(); it++)
+        if (isOutPin)
+            return ld.getLayerInstance()->outputNameToIndex(outName);
+        else
+            return ld.getLayerInstance()->inputNameToIndex(outName);
+    }
+
+    LayerPin getPinByAlias(const String &pinAlias, bool isOutPin = true)
     {
-            LayerData& ld = it->second;
+        LayerPin pin;
+        String layerName, outName;
+        splitPin(pinAlias, layerName, outName);
+
+        pin.lid = (layerName.empty()) ? 0 : getLayerId(layerName);
+
+        if (pin.lid >= 0)
+            pin.oid = resolvePinOutputName(getLayerData(pin.lid), outName, isOutPin);
 
-            std::cout << ld.name << std::endl;
-            std::cout << "Connected:" << 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;
+        return pin;
     }
+
+    void connect(int outLayerId, int outNum, int inLayerId, int inNum)
+    {
+        LayerData &ldOut = getLayerData(outLayerId);
+        LayerData &ldInp = getLayerData(inLayerId);
+
+        addLayerInput(ldInp, inNum, LayerPin(outLayerId, outNum));
+        ldOut.requiredOutputs.insert(outNum);
     }
 
-    void computeNetOutputs()
+    void computeNetOutputLayers()
     {
         netOutputs.clear();
 
@@ -351,31 +299,18 @@ struct Net::Impl
         for (set<int>::iterator i = ld.inputLayersId.begin(); i != ld.inputLayersId.end(); i++)
             allocateLayer(*i);
 
-        //create instance
-        if (ld.layerInstance == NULL && lid != 0)
-        {
-            ld.layerInstance = LayerRegister::createLayerInstance(ld.type, ld.params);
-            if (ld.layerInstance == NULL)
-            {
-                std::cerr << "Can't create layer \"" << ld.name << "\" of type \"" << ld.type << "\"" << std::endl;
-            }
-        }
-
         //bind inputs
         ld.inputBlobs.resize(ld.inputBlobsId.size());
         for (size_t i = 0; i < ld.inputBlobsId.size(); i++)
         {
-            int srcLId = ld.inputBlobsId[i].lid;
-            int srcOId = ld.inputBlobsId[i].oid;
-            ld.inputBlobs[i] = &layers[srcLId].outputBlobs[srcOId];
+            LayerPin from = ld.inputBlobsId[i];
+            CV_Assert(from.valid());
+            ld.inputBlobs[i] = &layers[from.lid].outputBlobs[from.oid];
         }
 
         //allocate layer
-        ld.outputBlobs.resize(ld.outputNames.size());
-        if (ld.layerInstance)
-            ld.layerInstance->allocate(ld.inputBlobs, ld.outputBlobs);
-
-        //std::cout << ld.name << " shape:" << ld.outputBlobs[0].shape() << std::endl;
+        ld.outputBlobs.resize(std::max((size_t)1, ld.requiredOutputs.size())); //layer produce at least one output blob
+        ld.getLayerInstance()->allocate(ld.inputBlobs, ld.outputBlobs);
 
         ld.flag = 1;
     }
@@ -393,7 +328,7 @@ struct Net::Impl
         }
     }
 
-    void forwardLayer(int layerId, bool clearFlags = true)
+    void forwardLayer(LayerData &ld, bool clearFlags = true)
     {
         if (clearFlags)
         {
@@ -402,8 +337,6 @@ struct Net::Impl
                 it->second.flag = 0;
         }
 
-        LayerData &ld = layers[layerId];
-
         //already was forwarded
         if (ld.flag)
             return;
@@ -411,15 +344,12 @@ struct Net::Impl
         //forward parents
         for (set<int>::iterator i = ld.inputLayersId.begin(); i != ld.inputLayersId.end(); i++)
         {
-            forwardLayer(*i, false);
+            forwardLayer(layers[*i], false);
         }
 
         //forward itself
-        if (ld.layerInstance && layerId != 0)
         ld.layerInstance->forward(ld.inputBlobs, ld.outputBlobs);
 
-        //std::cout << ld.name << " shape:" << ld.outputBlobs[0].shape() << std::endl;
-
         ld.flag = 1;
     }
 
@@ -430,7 +360,7 @@ struct Net::Impl
             it->second.flag = 0;
 
         for (it = layers.begin(); it != layers.end(); it++)
-            forwardLayer(it->first, false);
+            forwardLayer(it->second, false);
     }
 };
 
@@ -446,35 +376,38 @@ Net::~Net()
 
 int Net::addLayer(const String &name, const String &type, LayerParams &params)
 {
+    if (name.find('.') != String::npos)
+    {
+        CV_Error(Error::StsBadArg, "Added layer name \"" + name + "\" should not contain dot symbol");
+        return -1;
+    }
+
     if (impl->getLayerId(name) >= 0)
     {
-        CV_Error(cv::Error::StsBadArg, "Layer \"" + name + "\" already into net");
+        CV_Error(Error::StsBadArg, "Layer \"" + name + "\" already into net");
         return -1;
     }
 
     int id = ++impl->lastLayerId;
     impl->layerNameToId.insert(std::make_pair(name, id));
-    impl->layers.insert(std::make_pair(id, LayerData(name, type, params)));
+    impl->layers.insert(std::make_pair(id, LayerData(id, name, type, params)));
 
     return id;
 }
 
-//void Net::connect(BlobId input, BlobId output)
-//{
-
-//}
-
-void Net::setOutputNames(LayerId layer, const std::vector<String> &outputNames)
+void Net::connect(int outLayerId, int outNum, int inLayerId, int inNum)
 {
-    LayerData &ld = impl->getLayerData(layer);
-    CV_Assert(ld.outputNames.size() == 0);
-    ld.outputNames.assign(outputNames.begin(), outputNames.end());
+    impl->connect(outLayerId, outNum, inLayerId, inNum);
 }
 
-void Net::setLayerInputs(const std::vector<String> &outputs, LayerId layer)
+void Net::connect(String _outPin, String _inPin)
 {
-    LayerData &ld = impl->getLayerData(layer);
-    ld.inputNames.assign(outputs.begin(), outputs.end());
+    LayerPin outPin = impl->getPinByAlias(_outPin);
+    LayerPin inpPin = impl->getPinByAlias(_inPin);
+
+    CV_Assert(outPin.valid() && inpPin.valid());
+
+    impl->connect(outPin.lid, outPin.oid, inpPin.lid, inpPin.oid);
 }
 
 void Net::forward()
@@ -486,74 +419,62 @@ void Net::forward()
 void Net::forward(LayerId toLayer)
 {
     impl->setUpNet();
-    impl->forwardLayer(impl->getLayerId(toLayer));
+    impl->forwardLayer(impl->getLayerData(toLayer));
 }
 
 void Net::setNetInputs(const std::vector<String> &inputBlobNames)
 {
-    setOutputNames(0, inputBlobNames);
+    impl->netInputLayer->setNames(inputBlobNames);
 }
 
-void Net::setBlob(BlobId outputName, const Blob &blob)
+void Net::setBlob(String outputName, const Blob &blob)
 {
-    String name = outputName.get<String>();
-    LayerOutId found;
-
-    if (!impl->findOutputsByName(name, &found, 1))
-        CV_Error(cv::Error::StsObjectNotFound, "Request blob \"" + name + "\" not found");
+    LayerPin pin = impl->getPinByAlias(outputName);
+    if (!pin.valid())
+        CV_Error(Error::StsObjectNotFound, "Requested blob \"" + outputName + "\" not found");
 
-    LayerData &ld = impl->layers[found.lid];
-    ld.outputBlobs.resize(ld.outputNames.size());
-    ld.outputBlobs[found.oid] = blob;
+    LayerData &ld = impl->layers[pin.lid];
+    ld.outputBlobs.resize( std::max(pin.oid+1, (int)ld.requiredOutputs.size()) );
+    ld.outputBlobs[pin.oid] = blob;
 }
 
-Blob Net::getBlob(BlobId outputName)
+Blob Net::getBlob(String outputName)
 {
-    String name = outputName.get<String>();
-    LayerOutId found;
+    LayerPin pin = impl->getPinByAlias(outputName);
+    if (!pin.valid())
+        CV_Error(Error::StsObjectNotFound, "Requested blob \"" + outputName + "\" not found");
 
-    if (!impl->findOutputsByName(name, &found, 1))
-        CV_Error(cv::Error::StsObjectNotFound, "Request blob \"" + name + "\" not found");
-
-    LayerData &ld = impl->layers[found.lid];
-    return ld.outputBlobs[found.oid];
+    LayerData &ld = impl->layers[pin.lid];
+    if ((size_t)pin.oid >= ld.outputBlobs.size())
+    {
+        CV_Error(Error::StsOutOfRange, "Layer \"" + ld.name + "\" produce only " + toString(ld.outputBlobs.size()) +
+                                       " outputs, the #" + toString(pin.oid) + " was requsted");
+    }
+    return ld.outputBlobs[pin.oid];
 }
 
-Importer::~Importer()
+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];
 }
 
-//////////////////////////////////////////////////////////////////////////
-
-#include <sstream>
-template<typename T>
-String toString(const T &v)
-{
-    std::stringstream ss;
-    ss << v;
-    return ss.str();
-}
-
-int Layer::getNumInputs()
+Importer::~Importer()
 {
-    return 1;
-}
 
-int Layer::getNumOutputs()
-{
-    return 1;
 }
 
-cv::String Layer::getInputName(int inputNum)
+int Layer::inputNameToIndex(String)
 {
-    return "input" + toString(inputNum);
+    return -1;
 }
 
-
-cv::String Layer::getOutputName(int outputNum)
+int Layer::outputNameToIndex(String)
 {
-    return "output" + toString(outputNum);
+    return -1;
 }
 
 Layer::~Layer()
@@ -604,5 +525,11 @@ Ptr<Layer> LayerRegister::createLayerInstance(const String &_type, LayerParams&
     }
 }
 
+int Net::getLayerId(LayerId)
+{
+    CV_Error(Error::StsNotImplemented, "");
+    return -1;
+}
+
 }
 }

modules/dnn/src/layers/blank_layer.cpp

@@ -18,7 +18,7 @@ namespace dnn
         {
             outputs.resize(inputs.size());
             for (size_t i = 0; i < inputs.size(); i++)
-                outputs[i] = *inputs[i];
+                outputs[i].shareFrom(*inputs[i]);
         }
 
         void forward(std::vector<Blob*> &inputs, std::vector<Blob> &outputs)
@@ -28,12 +28,6 @@ namespace dnn
         }
     };
 
-    static Ptr<Layer> blankLayerRegisterer(LayerParams &params)
-    {
-        return Ptr<Layer>(new BlankLayer(params));
-    }
-
-
-    REGISTER_LAYER_FUNC(Dropout, blankLayerRegisterer)
+    REGISTER_LAYER_CLASS(Dropout, BlankLayer)
 }
 }

modules/dnn/src/layers/concat_layer.cpp

@@ -0,0 +1,73 @@
+#include "../precomp.hpp"
+#include "layers_common.hpp"
+
+namespace cv
+{
+namespace dnn
+{
+    class ConcatLayer : public Layer
+    {
+        int axis;
+
+    public:
+        ConcatLayer(LayerParams& params);
+        void allocate(const std::vector<Blob*> &inputs, std::vector<Blob> &outputs);
+        void forward(std::vector<Blob*> &inputs, std::vector<Blob> &outputs);
+    };
+
+
+    REGISTER_LAYER_CLASS(Concat, ConcatLayer)
+
+
+    ConcatLayer::ConcatLayer(LayerParams &params)
+    {
+        axis = params.get<int>("axis", 1);
+        CV_Assert(axis >= 0);
+    }
+
+    void ConcatLayer::allocate(const std::vector<Blob *> &inputs, std::vector<Blob> &outputs)
+    {
+        CV_Assert(inputs.size() > 0);
+
+        int refType = inputs[0]->type();
+        BlobShape refShape = inputs[0]->shape();
+        CV_Assert(axis < refShape.dims());
+
+        int axisSum = 0;
+        for (size_t i = 0; i < inputs.size(); i++)
+        {
+            BlobShape curShape = inputs[i]->shape();
+
+            CV_Assert(curShape.dims() == refShape.dims() && inputs[i]->type() == refType);
+            for (int axisId = 0; axisId < refShape.dims(); axisId++)
+            {
+                if (axisId != axis && refShape[axisId] != curShape[axisId])
+                    CV_Error(Error::StsBadSize, "Inconsitent shape for ConcatLayer");
+            }
+
+            axisSum += curShape[axis];
+        }
+
+        refShape[axis] = axisSum;
+        outputs.resize(1);
+        outputs[0].create(refShape);
+    }
+
+    void ConcatLayer::forward(std::vector<Blob *> &inputs, std::vector<Blob> &outputs)
+    {
+        const Mat& outMat = outputs[0].getMatRef();
+        std::vector<Range> ranges(outputs[0].dims(), Range::all());
+        int sizeStart = 0;
+        for (size_t i = 0; i < inputs.size(); i++)
+        {
+            int sizeEnd = sizeStart + inputs[i]->size(axis);
+            ranges[axis] = Range(sizeStart, sizeEnd);
+
+            Mat outSubMat = outMat(&ranges[0]);
+            inputs[i]->getMatRef().copyTo(outSubMat);
+
+            sizeStart = sizeEnd;
+        }
+    }
+}
+}

modules/dnn/src/layers/convolution_layer.cpp

@@ -1,5 +1,6 @@
 #include "../precomp.hpp"
 #include "layers_common.hpp"
+#include "im2col.hpp"
 
 namespace cv
 {
@@ -8,19 +9,24 @@ namespace dnn
     //TODO: simultaneously convolution and bias addition for cache optimization
     class ConvolutionLayer : public Layer
     {
+    protected:
         bool bias;
         int numOutput, group;
         int padH, padW;
+        int kerH, kerW;
         int strideH, strideW;
-        int kernelH, kernelW;
 
-        int inH, inW, inCn, kerSize;
-        int outH, outW;
-        int groupCn, groupCnOut;
+        int inpH, inpW, inpCn;
+        int outH, outW, outCn;
+        int topH, topW, topCn; //switched between inp/out on deconv/conv
+        int inpGroupCn, outGroupCn;
+        int ksize;
 
-        Mat srcColsMat, biasOnesMat;
+        Mat colMat, biasOnesMat;
 
-        void computeOutputShape(int inH, int inW);
+        inline bool is1x1() const;
+        virtual void computeInpOutShape(const Blob &inpBlob);
+        void im2col(Blob &inpBlob, int imNum, int cnGroup);
 
     public:
         ConvolutionLayer(LayerParams &params);
@@ -28,13 +34,25 @@ namespace dnn
         void forward(std::vector<Blob*> &inputs, std::vector<Blob> &outputs);
     };
 
+    class DeConvolutionLayer : public ConvolutionLayer
+    {
+    protected:
+        void computeInpOutShape(const Blob &inpBlob);
+        void col2im(Mat &dstMat);
+
+    public:
+        DeConvolutionLayer(LayerParams &params) : ConvolutionLayer(params) {}
+        void forward(std::vector<Blob*> &inputs, std::vector<Blob> &outputs);
+    };
+
 
     REGISTER_LAYER_CLASS(Convolution, ConvolutionLayer)
+    REGISTER_LAYER_CLASS(Deconvolution, DeConvolutionLayer)
 
 
     ConvolutionLayer::ConvolutionLayer(LayerParams &params)
     {
-        getKernelParams(params, kernelH, kernelW, padH, padW, strideH, strideW);
+        getKernelParams(params, kerH, kerW, padH, padW, strideH, strideW);
 
         numOutput = params.get<int>("num_output");
         bias = params.get<bool>("bias_term", true);
@@ -44,8 +62,8 @@ namespace dnn
         CV_Assert(params.learnedBlobs.size() >= 1 && (!bias || params.learnedBlobs.size() >= 2));
         learnedParams.assign(params.learnedBlobs.begin(), params.learnedBlobs.begin() + (bias ? 2 : 1));
 
-        Blob &weightBlob = learnedParams[0];
-        CV_Assert(weightBlob.cols() == kernelW && weightBlob.rows() == kernelH && weightBlob.num() == numOutput);
+        const Blob &wgtBlob = learnedParams[0];
+        CV_Assert(wgtBlob.dims() == 4 && wgtBlob.cols() == kerW && wgtBlob.rows() == kerH && wgtBlob.num() == numOutput);
 
         if (bias)
         {
@@ -58,94 +76,141 @@ namespace dnn
     {
         CV_Assert(inputs.size() > 0);
 
-        Blob &weightBlob = learnedParams[0];
+        const Blob &inpBlob = *inputs[0];
+        CV_Assert(inpBlob.dims() == 4 && inpBlob.type() == CV_32F);
+        computeInpOutShape(inpBlob);
 
-        inCn = inputs[0]->channels();
-        CV_Assert(inCn % group == 0 && numOutput % group == 0 && weightBlob.channels() == inCn/group);
-        groupCnOut = numOutput / group;
-        groupCn = inCn / group;
+        CV_Assert(inpCn % group == 0 && outCn % group == 0);
+        CV_Assert(learnedParams[0].channels() == inpCn / group);
+        CV_Assert(learnedParams[0].num() == outCn);
 
-        inH = inputs[0]->rows();
-        inW = inputs[0]->cols();
-        computeOutputShape(inH, inW);
+        outGroupCn = outCn / group;
+        inpGroupCn = inpCn / group;
+        ksize = inpGroupCn * kerH * kerW;
 
         outputs.resize(inputs.size());
         for (size_t i = 0; i < inputs.size(); i++)
         {
-            CV_Assert(inputs[i]->rows() == inH && inputs[i]->cols() == inW && inputs[i]->channels() == inCn);
-            int num = inputs[i]->num();
+            CV_Assert(inputs[i]->type() == inpBlob.type());
+            CV_Assert(inputs[i]->dims() == 4 && inputs[i]->channels() == inpBlob.channels());
+            CV_Assert(inputs[i]->rows() == inpBlob.rows() && inputs[i]->cols() == inpBlob.cols());
 
-            outputs[i].create(num, numOutput, outH, outW);
+            outputs[i].create(BlobShape(inputs[i]->num(), topCn, topH, topW));
         }
 
-        kerSize = kernelH * kernelW * groupCn;
-        srcColsMat.create(kerSize, outH * outW, CV_32F);
+        if (!is1x1())
+            colMat.create(ksize, outH * outW, inpBlob.type());
 
         if (bias)
-        {
-            biasOnesMat = Mat::ones(1, outH * outW, CV_32F);
+            biasOnesMat = Mat::ones(1, topH * topW, inpBlob.type());
     }
+
+    inline bool ConvolutionLayer::is1x1() const
+    {
+        return (kerH == 1 && kerW == 1);
     }
 
-    template <typename Dtype>
-    void im2col_cpu(const Dtype* data_im, const int channels,
-        const int height, const int width, const int kernel_h, const int kernel_w,
-        const int pad_h, const int pad_w,
-        const int stride_h, const int stride_w,
-        Dtype* data_col)
+    void ConvolutionLayer::forward(std::vector<Blob*> &inputs, std::vector<Blob> &outputs)
+    {
+        Blob &wgtBlob = learnedParams[0];
+
+        for (size_t ii = 0; ii < outputs.size(); ii++)
+        {
+            Blob &inpBlob = *inputs[ii];
+            Blob &outBlob = outputs[ii];
+
+            for (int n = 0; n < inpBlob.num(); n++)
+            {
+                for (int g = 0; g < group; g++)
+                {
+                    im2col(inpBlob, n, g);
+
+                    Mat kerMat(outGroupCn, ksize, wgtBlob.type(), wgtBlob.ptrRaw(g*outGroupCn));
+                    Mat dstMat(outGroupCn, outH*outW, outBlob.type(), outBlob.ptrRaw(n, g*outGroupCn));
+
+                    cv::gemm(kerMat, colMat, 1, noArray(), 0, dstMat);
+
+                    if (bias)
                     {
-        int height_col = (height + 2 * pad_h - kernel_h) / stride_h + 1;
-        int width_col = (width + 2 * pad_w - kernel_w) / stride_w + 1;
-        int channels_col = channels * kernel_h * kernel_w;
-        for (int c = 0; c < channels_col; ++c) {
-            int w_offset = c % kernel_w;
-            int h_offset = (c / kernel_w) % kernel_h;
-            int c_im = c / kernel_h / kernel_w;
-            for (int h = 0; h < height_col; ++h) {
-                for (int w = 0; w < width_col; ++w) {
-                    int h_pad = h * stride_h - pad_h + h_offset;
-                    int w_pad = w * stride_w - pad_w + w_offset;
-                    if (h_pad >= 0 && h_pad < height && w_pad >= 0 && w_pad < width)
-                        data_col[(c * height_col + h) * width_col + w] =
-                        data_im[(c_im * height + h_pad) * width + w_pad];
-                    else
-                        data_col[(c * height_col + h) * width_col + w] = 0;
+                        float *biasPtr = learnedParams[1].ptrf() + g*outGroupCn;
+                        Mat biasMat(outGroupCn, 1, CV_32F, biasPtr);
+                        cv::gemm(biasMat, biasOnesMat, 1, dstMat, 1, dstMat);
+                    }
                 }
             }
         }
     }
 
-    void ConvolutionLayer::forward(std::vector<Blob*> &inputs, std::vector<Blob> &outputs)
+    void ConvolutionLayer::im2col(Blob &inpBlob, int imNum, int cnGroup)
+    {
+        uchar *srcPtr = inpBlob.ptrRaw(imNum, cnGroup*inpGroupCn);
+
+        if (is1x1())
+        {
+            colMat = Mat(ksize, inpBlob.rows()*inpBlob.cols(), inpBlob.type(), srcPtr);
+            return;
+        }
+
+        if (inpBlob.type() == CV_32F)
+            im2col_cpu((float *)srcPtr, inpGroupCn, inpH, inpW, kerH, kerW, padH, padW, strideH, strideW, (float *)colMat.ptr());
+        if (inpBlob.type() == CV_64F)
+            im2col_cpu((double*)srcPtr, inpGroupCn, inpH, inpW, kerH, kerW, padH, padW, strideH, strideW, (double*)colMat.ptr());
+    }
+
+    void ConvolutionLayer::computeInpOutShape(const Blob &inpBlob)
     {
-        CV_Assert(inputs.size() == outputs.size());
+        inpH = inpBlob.rows();
+        inpW = inpBlob.cols();
+        inpCn = inpBlob.channels();
 
-        float *srcColPtr = srcColsMat.ptr<float>();
+        outH = (inpH + 2 * padH - kerH) / strideH + 1;
+        outW = (inpW + 2 * padW - kerW) / strideW + 1;
+        outCn = learnedParams[0].num();
+
+        topH = outH; topW = outW; topCn = outCn;
+    }
+
+    void DeConvolutionLayer::computeInpOutShape(const Blob &inpBlob)
+    {
+        outH = inpBlob.rows();
+        outW = inpBlob.cols();
+        outCn = inpBlob.channels();
+
+        inpH = strideH * (outH - 1) + kerH - 2 * padH;
+        inpW = strideW * (outW - 1) + kerW - 2 * padW;
+        inpCn = learnedParams[0].channels();
+
+        topH = inpH; topW = inpW; topCn = inpCn;
+    }
+
+    void DeConvolutionLayer::forward(std::vector<Blob*> &inputs, std::vector<Blob> &outputs)
+    {
+        Blob &wghtBlob = learnedParams[0];
 
         for (size_t ii = 0; ii < outputs.size(); ii++)
         {
-            Blob &input = *inputs[ii];
-            Blob &output = outputs[ii];
-            int num = input.num();
+            Blob &convBlob = *inputs[ii];
+            Blob &decnBlob = outputs[ii];
 
-            for (int n = 0; n < num; n++)
+            for (int n = 0; n < convBlob.num(); n++)
             {
                 for (int g = 0; g < group; g++)
                 {
-                    float *srcPtr = input.ptr<float>(n, g*groupCn);
-                    im2col_cpu(srcPtr, groupCn, inH, inW, kernelH, kernelW, padH, padW, strideH, strideW, srcColPtr);
+                    Mat dstMat(inpGroupCn, inpH*inpW, decnBlob.type(), decnBlob.ptrRaw(n, g*inpGroupCn));
 
-                    float *kerPtr = learnedParams[0].ptr<float>(g*groupCnOut);
-                    float *dstPtr = output.ptr<float>(n, g*groupCnOut);
+                    if (is1x1())
+                        colMat = dstMat;
 
-                    Mat kerMat(groupCnOut, kerSize, CV_32F, kerPtr);
-                    Mat dstMat(groupCnOut, outH*outW, CV_32F, dstPtr);
+                    Mat convMat(outGroupCn, outH*outW, convBlob.type(), convBlob.ptrRaw(n, g*inpGroupCn));
+                    Mat wghtMat(outGroupCn, ksize, wghtBlob.type(), wghtBlob.ptrRaw(g*inpGroupCn));
+                    cv::gemm(wghtMat, convMat, 1, noArray(), 0, colMat, GEMM_1_T);
 
-                    cv::gemm(kerMat, srcColsMat, 1, noArray(), 0, dstMat);
+                    col2im(dstMat);
 
                     if (bias)
                     {
-                        float *biasPtr = learnedParams[1].ptr<float>() + g*groupCnOut;
-                        Mat biasMat(groupCnOut, 1, CV_32F, biasPtr);
+                        float *biasPtr = learnedParams[1].ptrf() + g*outGroupCn;
+                        Mat biasMat(outGroupCn, 1, CV_32F, biasPtr);
                         cv::gemm(biasMat, biasOnesMat, 1, dstMat, 1, dstMat);
                     }
                 }
@@ -153,10 +218,14 @@ namespace dnn
         }
     }
 
-    void ConvolutionLayer::computeOutputShape(int inH, int inW)
+    void DeConvolutionLayer::col2im(Mat &dstMat)
     {
-        outH = (inH + 2 * padH - kernelH) / strideH + 1;
-        outW = (inW + 2 * padW - kernelW) / strideW + 1;
+        if (is1x1()) return;
+
+        if (dstMat.type() == CV_32F)
+            col2im_cpu((float*)colMat.ptr(), inpCn, inpH, inpW, kerH, kerW, padH, padW, strideH, strideW, (float*)dstMat.ptr());
+        if (dstMat.type() == CV_64F)
+            col2im_cpu((double*)colMat.ptr(), inpCn, inpH, inpW, kerH, kerW, padH, padW, strideH, strideW, (double*)dstMat.ptr());
     }
 }
 }

modules/dnn/src/layers/elementwise_layers.cpp

@@ -1,6 +1,10 @@
 #include "../precomp.hpp"
 #include "layers_common.hpp"
-#include <math.h>
+#include <cmath>
+using std::abs;
+using std::exp;
+using std::tanh;
+using std::pow;
 
 namespace cv
 {
@@ -19,22 +23,34 @@ namespace dnn
         {
             outputs.resize(inputs.size());
             for (size_t i = 0; i < inputs.size(); i++)
-                outputs[i] = *inputs[i]; //no data copy
+                outputs[i].shareFrom(*inputs[i]); //no data copy
         }
 
         void forward(std::vector<Blob*> &inputs, std::vector<Blob> &outputs)
         {
-            CV_Assert(inputs.size() == outputs.size());
-
             for (size_t i = 0; i < inputs.size(); i++)
             {
-                CV_Assert(inputs[i]->ptr<float>() == outputs[i].ptr<float>());
-                float *data = outputs[i].ptr<float>();
+                CV_Assert(inputs[i]->ptrRaw() == outputs[i].ptrRaw() && inputs[i]->type() == outputs[i].type());
+
                 size_t size = outputs[i].total();
 
+                if (outputs[i].isFloat())
+                {
+                    float *data = outputs[i].ptrf();
+                    for (size_t j = 0; j < size; j++)
+                        data[j] = func(data[j]);
+                }
+                else if (outputs[i].isDouble())
+                {
+                    double *data = outputs[i].ptr<double>();
                     for (size_t j = 0; j < size; j++)
                         data[j] = func(data[j]);
                 }
+                else
+                {
+                    CV_Error(Error::StsNotImplemented, "Only CV_32F and CV_64F blobs are supported");
+                }
+            }
         }
     };
 
@@ -51,9 +67,10 @@ namespace dnn
                 negative_slope = 0.f;
         }
 
-        inline float operator()(float x)
+        template<typename TFloat>
+        inline TFloat operator()(TFloat x)
         {
-            return (x >= 0) ? x : negative_slope * x;
+            return (x >= (TFloat)0) ? x : negative_slope * x;
         }
     };
 
@@ -61,14 +78,70 @@ namespace dnn
     {
         TanHFunctor(LayerParams&) {}
 
-        inline float operator()(float x)
+        template<typename TFloat>
+        inline TFloat operator()(TFloat x)
         {
             return tanh(x);
         }
     };
 
+    struct SigmoidFunctor
+    {
+        SigmoidFunctor(LayerParams&) {}
+
+        template<typename TFloat>
+        inline TFloat operator()(TFloat x)
+        {
+            return (TFloat)1 / ((TFloat)1 + exp(-x));
+        }
+    };
+
+    struct AbsValFunctor
+    {
+        AbsValFunctor(LayerParams&) {}
+
+        template<typename TFloat>
+        inline TFloat operator()(TFloat x)
+        {
+            return abs(x);
+        }
+    };
+
+    struct PowerFunctor
+    {
+        float power, scale, shift;
+
+        PowerFunctor(LayerParams &params)
+        {
+            power = params.get<float>("power", 1.0f);
+            scale = params.get<float>("scale", 1.0f);
+            shift = params.get<float>("shift", 0.0f);
+        }
+
+        template<typename TFloat>
+        inline TFloat operator()(TFloat x)
+        {
+            return pow((TFloat)shift + (TFloat)scale * x, (TFloat)power);
+        }
+    };
+
+    struct BNLLFunctor
+    {
+        BNLLFunctor(LayerParams&) {}
+
+        template<typename TFloat>
+        inline TFloat operator()(TFloat x)
+        {
+            return log((TFloat)1 + exp(-abs(x)));
+        }
+    };
+
     REGISTER_LAYER_CLASS(ReLU, ElementWiseLayer<ReLUFunctor>)
     REGISTER_LAYER_CLASS(TanH, ElementWiseLayer<TanHFunctor>)
+    REGISTER_LAYER_CLASS(BNLL, ElementWiseLayer<BNLLFunctor>)
+    REGISTER_LAYER_CLASS(Power, ElementWiseLayer<PowerFunctor>)
+    REGISTER_LAYER_CLASS(AbsVal, ElementWiseLayer<AbsValFunctor>)
+    REGISTER_LAYER_CLASS(Sigmoid, ElementWiseLayer<SigmoidFunctor>)
 
 }
 }

modules/dnn/src/layers/fully_connected_layer.cpp

@@ -1,18 +1,20 @@
 #include "../precomp.hpp"
 #include "layers_common.hpp"
+#include <iostream>
 
 namespace cv
 {
 namespace dnn
 {
-    //TODO: implement axis number parameter
     class FullyConnectedLayer : public Layer
     {
         bool bias;
         int numOutputs;
+        int axis_, axis;
 
-        int inC, inH, inW;
-        int inSize;
+        int innerSize;
+
+        void reshape(const Blob &inp, Blob &out);
 
     public:
         FullyConnectedLayer(LayerParams &params);
@@ -28,6 +30,7 @@ namespace dnn
     {
         numOutputs = params.get<int>("num_output");
         bias = params.get<bool>("bias_term", true);
+        axis_ = params.get<int>("axis", 1);
 
         CV_Assert(params.learnedBlobs.size() >= 1);
         CV_Assert(!bias || (params.learnedBlobs.size() >= 2 && (int)params.learnedBlobs[1].total() == numOutputs));
@@ -44,41 +47,50 @@ namespace dnn
     {
         CV_Assert(inputs.size() > 0);
 
-        inC = inputs[0]->channels();
-        inH = inputs[0]->rows();
-        inW = inputs[0]->cols();
-        inSize = inC * inH * inW;
+        axis = inputs[0]->canonicalAxis(axis_);
+        innerSize = (int)inputs[0]->total(axis);
 
-        CV_Assert((size_t)inSize * (size_t)numOutputs == learnedParams[0].total());
+        CV_Assert((size_t)innerSize * (size_t)numOutputs == learnedParams[0].total());
+        CV_Assert(learnedParams[0].rows() == numOutputs && learnedParams[0].cols() == innerSize);
 
         outputs.resize(inputs.size());
         for (size_t i = 0; i < inputs.size(); i++)
         {
             if (i != 0)
-                CV_Assert(inputs[i]->channels() == inC && inputs[i]->rows() == inH && inputs[i]->cols() == inW);
+                CV_Assert(inputs[i]->equalShape(*inputs[0]));
 
-            outputs[i].create(inputs[i]->num(), numOutputs, 1, 1);
+            this->reshape(*inputs[i], outputs[i]);
+        }
     }
+
+    void FullyConnectedLayer::reshape(const Blob &inp, Blob &out)
+    {
+        BlobShape inpShape = inp.shape();
+        BlobShape outShape(axis+1, inpShape.ptr());
+        outShape[axis] = numOutputs;
+
+        out.create(outShape, inp.type());
     }
 
     void FullyConnectedLayer::forward(std::vector<Blob*> &inputs, std::vector<Blob> &outputs)
     {
         for (size_t i = 0; i < inputs.size(); i++)
         {
-            int M = inputs[i]->num();
+            int M = (int)inputs[i]->total(0, axis);
             int N = numOutputs;
-            int K = inSize;
+            int K = innerSize;
 
-            Mat srcMat(M, K, CV_32F, inputs[i]->ptr<float>());
-            Mat weights(K, N, CV_32F, learnedParams[0].ptr<float>());
-            Mat dstMat(M, N, CV_32F, outputs[i].ptr<float>());
+            Mat srcMat(M, K, inputs[i]->type(), inputs[i]->ptrf());
+            Mat weight(N, K, learnedParams[0].type(), learnedParams[0].ptrf());
+            Mat dstMat(M, N, outputs[i].type(), outputs[i].ptrf());
 
-            cv::gemm(srcMat, weights, 1, noArray(), 0, dstMat);
+            //important: Caffe stores weights as transposed array
+            cv::gemm(srcMat, weight, 1, noArray(), 0, dstMat, GEMM_2_T);
 
             if (bias)
             {
                 Mat biasOnesMat = Mat::ones(M, 1, CV_32F);
-                Mat biasMat(1, N, CV_32F, learnedParams[1].ptr<float>());
+                Mat biasMat(1, N, CV_32F, learnedParams[1].ptrf());
                 cv::gemm(biasOnesMat, biasMat, 1, dstMat, 1, dstMat);
             }
         }

modules/dnn/src/layers/im2col.hpp

@@ -0,0 +1,74 @@
+#ifndef __OPENCV_DNN_LAYERS_IM2COL_HPP__
+#define __OPENCV_DNN_LAYERS_IM2COL_HPP__
+
+namespace cv
+{
+namespace dnn
+{
+
+template <typename Dtype>
+void im2col_cpu(const Dtype* data_im, const int channels,
+    const int height, const int width, const int kernel_h, const int kernel_w,
+    const int pad_h, const int pad_w,
+    const int stride_h, const int stride_w,
+    Dtype* data_col)
+{
+    int height_col = (height + 2 * pad_h - kernel_h) / stride_h + 1;
+    int width_col = (width + 2 * pad_w - kernel_w) / stride_w + 1;
+    int channels_col = channels * kernel_h * kernel_w;
+    for (int c = 0; c < channels_col; ++c) {
+        int w_offset = c % kernel_w;
+        int h_offset = (c / kernel_w) % kernel_h;
+        int c_im = c / kernel_h / kernel_w;
+        for (int h = 0; h < height_col; ++h) {
+            for (int w = 0; w < width_col; ++w) {
+                int h_pad = h * stride_h - pad_h + h_offset;
+                int w_pad = w * stride_w - pad_w + w_offset;
+                if (h_pad >= 0 && h_pad < height && w_pad >= 0 && w_pad < width)
+                    data_col[(c * height_col + h) * width_col + w] =
+                    data_im[(c_im * height + h_pad) * width + w_pad];
+                else
+                    data_col[(c * height_col + h) * width_col + w] = 0;
+            }
+        }
+    }
+}
+
+template <typename Dtype>
+void col2im_cpu(const Dtype* data_col, const int channels,
+    const int height, const int width, const int patch_h, const int patch_w,
+    const int pad_h, const int pad_w,
+    const int stride_h, const int stride_w,
+    Dtype* data_im)
+{
+    memset(data_im, 0, height * width * channels * sizeof(Dtype));
+
+    int height_col = (height + 2 * pad_h - patch_h) / stride_h + 1;
+    int width_col = (width + 2 * pad_w - patch_w) / stride_w + 1;
+    int channels_col = channels * patch_h * patch_w;
+
+    for (int c = 0; c < channels_col; ++c)
+    {
+        int w_offset = c % patch_w;
+        int h_offset = (c / patch_w) % patch_h;
+        int c_im = c / patch_h / patch_w;
+
+        for (int h = 0; h < height_col; ++h)
+        {
+            for (int w = 0; w < width_col; ++w)
+            {
+                int h_pad = h * stride_h - pad_h + h_offset;
+                int w_pad = w * stride_w - pad_w + w_offset;
+
+                if (h_pad >= 0 && h_pad < height && w_pad >= 0 && w_pad < width)
+                    data_im[(c_im * height + h_pad) * width + w_pad] +=
+                    data_col[(c * height_col + h) * width_col + w];
+            }
+        }
+    }
+}
+
+}
+}
+
+#endif

modules/dnn/src/layers/lrn_layer.cpp

@@ -1,6 +1,7 @@
 #include "../precomp.hpp"
 #include "layers_common.hpp"
 #include <opencv2/imgproc.hpp>
+#include <algorithm>
 
 namespace cv
 {
@@ -45,8 +46,8 @@ namespace dnn
             CV_Error(cv::Error::StsBadArg, "Unknown region type \"" + nrmType + "\"");
 
         size = params.get<int>("local_size", 5);
-        if (size % 2 != 1)
-            CV_Error(cv::Error::StsBadArg, "LRN layer only supports odd values for local_size");
+        if (size % 2 != 1 || size <= 0)
+            CV_Error(cv::Error::StsBadArg, "LRN layer supports only positive odd values for local_size");
 
         alpha = params.get<double>("alpha", 1);
         beta = params.get<double>("beta", 0.75);
@@ -57,10 +58,10 @@ namespace dnn
         CV_Assert(inputs.size() == 1);
         outputs.resize(1);
 
-        Vec4i shape = inputs[0]->shape();
+        Vec4i shape = inputs[0]->shape4();
         outputs[0].create(shape);
 
-        shape[1] = 1; //maybe make shape[0] = 1 too
+        shape[0] = 1; //maybe make shape[0] = 1 too
         bufBlob.create(shape);
     }
 
@@ -85,26 +86,37 @@ namespace dnn
 
     void LRNLayer::channelNoramlization(Blob &srcBlob, Blob &dstBlob)
     {
+        CV_DbgAssert(srcBlob.ptrRaw() != dstBlob.ptrRaw());
+
         int num = srcBlob.num();
         int channels = srcBlob.channels();
+        int ksize = (size - 1) / 2;
 
         for (int n = 0; n < num; n++)
         {
-            Mat buf = bufBlob.getMat(n, 0);
-            Mat accum = dstBlob.getMat(n, 0); //memory saving
+            Mat accum = dstBlob.getMat(n, channels-1); //trick for memory saving
             accum.setTo(0);
 
+            for (int cn = 0; cn < std::min(ksize, channels); cn++)
+                cv::accumulateSquare(srcBlob.getMat(n, cn), accum);
+
             for (int cn = 0; cn < channels; cn++)
             {
-                cv::accumulateSquare(srcBlob.getMat(n, cn), accum);
+                if (cn + ksize < channels)
+                {
+                    cv::accumulateSquare(srcBlob.getMat(n, cn + ksize), accum);
                 }
 
-            accum.convertTo(accum, accum.type(), alpha/channels, 1);
-            cv::pow(accum, beta, accum);
-
-            for (int cn = channels - 1; cn >= 0; cn--)
+                if (cn - ksize - 1 >= 0)
                 {
-                cv::divide(srcBlob.getMat(n, cn), accum, dstBlob.getMat(n, cn));
+                    Mat left = srcBlob.getMat(n, cn - ksize - 1);
+                    cv::subtract(accum, left.mul(left), accum); //subtractSquare
+                }
+
+                Mat dst = dstBlob.getMat(n, cn);
+                accum.convertTo(dst, dst.type(), alpha/size, 1);
+                cv::pow(dst, beta, dst);
+                cv::divide(srcBlob.getMat(n, cn), dst, dst);
             }
         }
     }

modules/dnn/src/layers/pooling_layer.cpp

@@ -27,6 +27,7 @@ namespace dnn
 
         void computeOutputShape(int inH, int inW);
         void maxPooling(Blob &input, Blob &output);
+        void avePooling(Blob &input, Blob &output);
 
     public:
         PoolingLayer(LayerParams &params);
@@ -64,15 +65,15 @@ namespace dnn
     {
         CV_Assert(inputs.size() > 0);
 
-        inH = inputs[0]->cols();
-        inW = inputs[0]->rows();
+        inW = inputs[0]->cols();
+        inH = inputs[0]->rows();
         computeOutputShape(inH, inW);
 
         outputs.resize(inputs.size());
         for (size_t i = 0; i < inputs.size(); i++)
         {
             CV_Assert(inputs[i]->rows() == inH && inputs[i]->cols() == inW);
-            outputs[i].create(inputs[i]->num(), inputs[i]->channels(), pooledH, pooledW);
+            outputs[i].create(BlobShape(inputs[i]->num(), inputs[i]->channels(), pooledH, pooledW));
         }
     }
 
@@ -85,6 +86,9 @@ namespace dnn
             case MAX:
                 maxPooling(*inputs[ii], outputs[ii]);
                 break;
+            case AVE:
+                avePooling(*inputs[ii], outputs[ii]);
+                break;
             default:
                 CV_Error(cv::Error::StsNotImplemented, "Not implemented");
                 break;
@@ -100,8 +104,8 @@ namespace dnn
         {
             for (int c = 0; c < input.channels(); ++c)
             {
-                float *srcData = input.ptr<float>(n, c);
-                float *dstData = output.ptr<float>(n, c);
+                float *srcData = input.ptrf(n, c);
+                float *dstData = output.ptrf(n, c);
 
                 for (int ph = 0; ph < pooledH; ++ph)
                 {
@@ -131,6 +135,42 @@ namespace dnn
         }
     }
 
+    void PoolingLayer::avePooling(Blob &input, Blob &output)
+    {
+        for (int n = 0; n < input.num(); ++n)
+        {
+            for (int c = 0; c < input.channels(); ++c)
+            {
+                float *srcData = input.ptrf(n, c);
+                float *dstData = output.ptrf(n, c);
+
+                for (int ph = 0; ph < pooledH; ++ph)
+                {
+                    for (int pw = 0; pw < pooledW; ++pw)
+                    {
+                        int hstart = ph * strideH - padH;
+                        int wstart = pw * strideH - padH;
+                        int hend = min(hstart + kernelH, inW + padH);
+                        int wend = min(wstart + kernelW, inH + padW);
+                        int pool_size = (hend - hstart) * (wend - wstart);
+                        hstart = max(hstart, 0);
+                        wstart = max(wstart, 0);
+                        hend = min(hend, inH);
+                        wend = min(wend, inW);
+
+                        dstData[ph * pooledH + pw] = 0.f;
+
+                        for (int h = hstart; h < hend; ++h)
+                            for (int w = wstart; w < wend; ++w)
+                                dstData[ph * pooledH + pw] += srcData[h * inW + w];
+
+                        dstData[ph * pooledH + pw] /= pool_size;
+                    }
+                }
+          }
+        }
+    }
+
     void PoolingLayer::computeOutputShape(int inH, int inW)
     {
         //Yeah, something strange Caffe scheme-)

modules/dnn/src/layers/reshape_layer.cpp

@@ -0,0 +1,137 @@
+#include "../precomp.hpp"
+#include "layers_common.hpp"
+
+namespace cv
+{
+namespace dnn
+{
+
+//TODO: Extend cv::Mat::reshape method
+class ReshapeLayer : public Layer
+{
+public:
+    ReshapeLayer(LayerParams &params);
+
+    void allocate(const std::vector<Blob*> &inputs, std::vector<Blob> &outputs);
+
+    void forward(std::vector<Blob*>&, std::vector<Blob>&) {}
+
+protected:
+    BlobShape shapeDesc;
+    int inAxis, inNumAxes, autoAxisIdx;
+
+    void computeOutputShape(int startAxis, int endAxis, BlobShape &inpShape, BlobShape &outShape);
+};
+
+ReshapeLayer::ReshapeLayer(LayerParams &params)
+{
+    DictValue paramShape = params.get("dim");
+    shapeDesc = BlobShape(paramShape.size());
+    autoAxisIdx = -1;
+
+    for (int i = 0; i < paramShape.size(); i++)
+    {
+        int dim = paramShape.get<int>(i);
+        CV_Assert(dim >= -1);
+
+        if (dim == -1)
+        {
+            if (autoAxisIdx != -1)
+                CV_Error(Error::StsBadArg, "New shape contains multiple -1 dims");
+            autoAxisIdx = i;
+        }
+
+        shapeDesc[i] = dim;
+    }
+
+    inAxis = params.get<int>("axis", 0);
+    inNumAxes = params.get<int>("num_axes", -1);
+    CV_Assert(inNumAxes >= -1);
+}
+
+void ReshapeLayer::allocate(const std::vector<Blob*> &inputs, std::vector<Blob> &outputs)
+{
+    CV_Assert(inputs.size() == 1);
+    outputs.resize(1);
+
+    Blob &inpBlob = *inputs[0];
+    Blob &outBlob = outputs[0];
+    BlobShape inpShape = inpBlob.shape();
+
+    int startAxis = (inAxis >= 0) ? inAxis : inpShape.dims() + 1 + inAxis;
+    int endAxis = (inNumAxes == -1) ? inpShape.dims() : startAxis + inNumAxes;
+    CV_Assert(0 <= startAxis && startAxis <= inpShape.dims());
+    CV_Assert(0 <= endAxis && endAxis <= inpShape.dims());
+
+    int newDims = inpShape.dims() - (endAxis - startAxis) + shapeDesc.dims();
+    BlobShape outShape(newDims);
+
+    computeOutputShape(startAxis, endAxis, inpShape, outShape);
+
+    outBlob.shareFrom(inpBlob);
+    outBlob.reshape(outShape);
+}
+
+void ReshapeLayer::computeOutputShape(int startAxis, int endAxis, BlobShape &inpShape, BlobShape &outShape)
+{
+    int idx = 0;
+    for (int i = 0; i < startAxis; i++)
+        outShape[idx++] = inpShape[i];
+
+    for (int i = 0; i < shapeDesc.dims(); i++)
+    {
+        if (shapeDesc[i] == 0)
+        {
+            int inpAxisIdx = startAxis + i;
+            if (inpAxisIdx < 0 || inpShape.dims() <= inpAxisIdx)
+                CV_Error(Error::StsOutOfRange, "new shape contains a 0, but there was no corresponding bottom axis to copy");
+            outShape[idx++] = inpShape[startAxis + i];
+        }
+        else
+        {
+            outShape[idx++] = (shapeDesc[i] > 0) ? shapeDesc[i] : 1;
+        }
+    }
+
+    for (int i = endAxis; i < inpShape.dims(); i++)
+        outShape[idx++] = inpShape[i];
+
+    if (autoAxisIdx >= 0)
+    {
+        size_t total = inpShape.total();
+        size_t curTotal = 1;
+        for (int i = 0; i < outShape.dims(); i++)
+        {
+            if (i != startAxis + autoAxisIdx)
+                curTotal *= outShape[i];
+        }
+
+        CV_DbgAssert(curTotal <= total && total % curTotal == 0);
+
+        outShape[startAxis + autoAxisIdx] = (int)(total / curTotal);
+    }
+
+    if (inpShape.total() != outShape.total())
+    {
+        CV_Error(Error::StsBadArg, "Mismatch between input and output blob elements count");
+    }
+}
+
+
+Ptr<Layer> createFlattenLayer(LayerParams&)
+{
+    LayerParams params;
+
+    int shapeDesc[] = {0, -1};
+    params.set("dim", DictValue::arrayInt(shapeDesc, 2));
+
+    return Ptr<Layer>(new ReshapeLayer(params));
+}
+
+
+REGISTER_LAYER_CLASS(Reshape, ReshapeLayer)
+REGISTER_LAYER_FUNC(Flatten, createFlattenLayer)
+
+
+}
+}

modules/dnn/src/layers/slice_layer.cpp

@@ -0,0 +1,103 @@
+#include "../precomp.hpp"
+#include "layers_common.hpp"
+
+namespace cv
+{
+namespace dnn
+{
+
+class SliceLayer : public Layer
+{
+public:
+    SliceLayer(LayerParams &params);
+
+    void allocate(const std::vector<Blob*> &inputs, std::vector<Blob> &outputs);
+
+    void forward(std::vector<Blob*> &inputs, std::vector<Blob> &outputs);
+
+private:
+    int inAxis;
+    std::vector<int> slicePoints;
+};
+
+
+REGISTER_LAYER_CLASS(Slice, SliceLayer)
+
+
+SliceLayer::SliceLayer(LayerParams &params)
+{
+    inAxis = params.get<int>("axis", 1);
+
+    const DictValue &_slicePoints = params.get("slice_point");
+    slicePoints.resize(_slicePoints.size());
+    for (int i = 0; i < _slicePoints.size(); i++)
+    {
+        slicePoints[i] = _slicePoints.get<int>(i);
+        CV_Assert(slicePoints[i] > 0 && (i == 0 || slicePoints[i-1] < slicePoints[i]));
+    }
+}
+
+void SliceLayer::allocate(const std::vector<Blob*> &inputs, std::vector<Blob> &outputs)
+{
+    CV_Assert(inputs.size() == 1);
+
+    const Blob inpBlob = *inputs[0];
+    int axis = inpBlob.canonicalAxis(inAxis);
+    int axisSize = inpBlob.size(axis);
+    BlobShape inpShape = inpBlob.shape();
+
+    if (slicePoints.size()) //divide blob with respect to passed parameters
+    {
+        std::vector<int> outAxisSize;
+        int prevSlice = 0;
+
+        for (size_t i = 0; i < slicePoints.size(); i++)
+        {
+            CV_Assert(prevSlice < slicePoints[i] && slicePoints[i] < axisSize);
+            outAxisSize.push_back(slicePoints[i] - prevSlice);
+            prevSlice = slicePoints[i];
+        }
+        outAxisSize.push_back(axisSize - prevSlice);
+
+        outputs.resize(outAxisSize.size());
+        for (size_t i = 0; i < outAxisSize.size(); i++)
+        {
+            inpShape[axis] = outAxisSize[i];
+            outputs[i].create(inpShape, inpBlob.type());
+        }
+    }
+    else //divide blob with respect to count of output blobs
+    {
+        CV_Assert(outputs.size() > 0 && axisSize % outputs.size() == 0);
+        int outAxisSize = axisSize / (int)outputs.size();
+
+        for (size_t i = 0; i < outputs.size(); i++)
+        {
+            inpShape[axis] = outAxisSize;
+            outputs[i].create(inpShape, inpBlob.type());
+        }
+    }
+}
+
+void SliceLayer::forward(std::vector<Blob*> &inputs, std::vector<Blob> &outputs)
+{
+    Blob &inpBlob = *inputs[0];
+    const int axis = inpBlob.canonicalAxis(inAxis);
+    const Mat& inpMat = inpBlob.getMatRef();
+
+    std::vector<Range> ranges(inpBlob.dims(), Range::all());
+    int sizeStart = 0;
+    for (size_t i = 0; i < outputs.size(); i++)
+    {
+        int sizeEnd = sizeStart + outputs[i].size(axis);
+        ranges[axis] = Range(sizeStart, sizeEnd);
+
+        Mat inpSubMat = inpMat(&ranges[0]);
+        inpSubMat.copyTo(outputs[i].getMatRef());
+
+        sizeStart = sizeEnd;
+    }
+}
+
+}
+}

modules/dnn/src/layers/softmax_layer.cpp

@@ -8,9 +8,10 @@ namespace cv
 {
 namespace dnn
 {
+    //TODO: set default axis number to 1, and add custom shape length in FullyConnected
     class SoftMaxLayer : public Layer
     {
-        int axis;
+        int axis_, axis;
         Blob maxAggregator;
 
     public:
@@ -25,15 +26,16 @@ namespace dnn
 
     SoftMaxLayer::SoftMaxLayer(LayerParams &params)
     {
-        axis = params.get<int>("axis", 1);
-        CV_Assert(0 <= axis && axis < 4);
+        //hotfix!!!
+        axis_ = params.get<int>("axis", 1);
     }
 
     void SoftMaxLayer::allocate(const std::vector<Blob*> &inputs, std::vector<Blob> &outputs)
     {
         CV_Assert(inputs.size() == 1);
+        axis = inputs[0]->canonicalAxis(axis_);
 
-        Vec4i shape = inputs[0]->shape();
+        BlobShape shape = inputs[0]->shape();
         outputs.resize(1);
         outputs[0].create(shape);
 
@@ -46,9 +48,9 @@ namespace dnn
         Blob &src = *inputs[0];
         Blob &dst = outputs[0];
 
-        float *srcPtr = src.ptr<float>();
-        float *dstPtr = dst.ptr<float>();
-        float *bufPtr = maxAggregator.ptr<float>();
+        float *srcPtr = src.ptrf();
+        float *dstPtr = dst.ptrf();
+        float *bufPtr = maxAggregator.ptrf();
 
         size_t outerSize = src.total(0, axis);
         size_t channels = src.size(axis);
@@ -85,7 +87,7 @@ namespace dnn
             }
         }
 
-        cv::exp(dst.getMat(), dst.getMat());
+        cv::exp(dst.getMatRef(), dst.getMatRef());
 
         for (size_t outerDim = 0; outerDim < outerSize; outerDim++)
         {

modules/dnn/src/layers/split_layer.cpp

@@ -0,0 +1,58 @@
+#include "../precomp.hpp"
+#include "layers_common.hpp"
+
+namespace cv
+{
+namespace dnn
+{
+
+//TODO: maybe "top_count" param is useless because it can be determined by output connections number?
+class SplitLayer : public Layer
+{
+public:
+    SplitLayer(LayerParams &params);
+
+    void allocate(const std::vector<Blob*> &inputs, std::vector<Blob> &outputs);
+
+    void forward(std::vector<Blob*> &inputs, std::vector<Blob> &outputs);
+
+private:
+    int outputsNum;
+};
+
+
+REGISTER_LAYER_CLASS(Split, SplitLayer)
+
+
+SplitLayer::SplitLayer(LayerParams &params)
+{
+    if (params.has("top_count"))
+    {
+        outputsNum = params.get<int>("top_count");
+        CV_Assert(outputsNum >= 0);
+    }
+    else
+    {
+        outputsNum = -1;
+    }
+}
+
+void SplitLayer::allocate(const std::vector<Blob*> &inputs, std::vector<Blob> &outputs)
+{
+    CV_Assert(inputs.size() == 1);
+
+    if (outputsNum >= 0)
+        outputs.resize(outputsNum);
+
+    for (size_t i = 0; i < outputs.size(); i++)
+        outputs[i].create(inputs[0]->shape(), inputs[0]->type());
+}
+
+void SplitLayer::forward(std::vector<Blob*> &inputs, std::vector<Blob> &outputs)
+{
+    for (size_t i = 0; i < outputs.size(); i++)
+        inputs[0]->getMatRef().copyTo(outputs[i].getMatRef());
+}
+
+}
+}

modules/dnn/test/cnpy.cpp

@@ -0,0 +1,247 @@
+//Copyright (C) 2011  Carl Rogers
+//Released under MIT License
+//license available in LICENSE file, or at http://www.opensource.org/licenses/mit-license.php
+
+#include"cnpy.h"
+#include<complex>
+#include<cstdlib>
+#include<algorithm>
+#include<cstring>
+#include<iomanip>
+
+char cnpy::BigEndianTest() {
+    unsigned char x[] = {1,0};
+    short y = *(short*) x;
+    return y == 1 ? '<' : '>';
+}
+
+char cnpy::map_type(const std::type_info& t)
+{
+    if(t == typeid(float) ) return 'f';
+    if(t == typeid(double) ) return 'f';
+    if(t == typeid(long double) ) return 'f';
+
+    if(t == typeid(int) ) return 'i';
+    if(t == typeid(char) ) return 'i';
+    if(t == typeid(short) ) return 'i';
+    if(t == typeid(long) ) return 'i';
+    if(t == typeid(long long) ) return 'i';
+
+    if(t == typeid(unsigned char) ) return 'u';
+    if(t == typeid(unsigned short) ) return 'u';
+    if(t == typeid(unsigned long) ) return 'u';
+    if(t == typeid(unsigned long long) ) return 'u';
+    if(t == typeid(unsigned int) ) return 'u';
+
+    if(t == typeid(bool) ) return 'b';
+
+    if(t == typeid(std::complex<float>) ) return 'c';
+    if(t == typeid(std::complex<double>) ) return 'c';
+    if(t == typeid(std::complex<long double>) ) return 'c';
+
+    else return '?';
+}
+
+template<> std::vector<char>& cnpy::operator+=(std::vector<char>& lhs, const std::string rhs) {
+    lhs.insert(lhs.end(),rhs.begin(),rhs.end());
+    return lhs;
+}
+
+template<> std::vector<char>& cnpy::operator+=(std::vector<char>& lhs, const char* rhs) {
+    //write in little endian
+    size_t len = strlen(rhs);
+    lhs.reserve(len);
+    for(size_t byte = 0; byte < len; byte++) {
+        lhs.push_back(rhs[byte]);
+    }
+    return lhs;
+}
+
+void cnpy::parse_npy_header(FILE* fp, unsigned int& word_size, unsigned int*& shape, unsigned int& ndims, bool& fortran_order) {
+    char buffer[256];
+    size_t res = fread(buffer,sizeof(char),11,fp);
+    if(res != 11)
+        throw std::runtime_error("parse_npy_header: failed fread");
+    std::string header = fgets(buffer,256,fp);
+    assert(header[header.size()-1] == '\n');
+
+    size_t loc1, loc2;
+
+    //fortran order
+    loc1 = header.find("fortran_order")+16;
+    fortran_order = (header.substr(loc1,5) == "True" ? true : false);
+
+    //shape
+    loc1 = header.find("(");
+    loc2 = header.find(")");
+    std::string str_shape = header.substr(loc1+1,loc2-loc1-1);
+    if(str_shape[str_shape.size()-1] == ',') ndims = 1;
+    else ndims = (unsigned)std::count(str_shape.begin(),str_shape.end(),',')+1;
+    shape = new unsigned int[ndims];
+    for(unsigned int i = 0;i < ndims;i++) {
+        loc1 = str_shape.find(",");
+        shape[i] = atoi(str_shape.substr(0,loc1).c_str());
+        str_shape = str_shape.substr(loc1+1);
+    }
+
+    //endian, word size, data type
+    //byte order code | stands for not applicable.
+    //not sure when this applies except for byte array
+    loc1 = header.find("descr")+9;
+    bool littleEndian = (header[loc1] == '<' || header[loc1] == '|' ? true : false);
+    assert(littleEndian);
+
+    //char type = header[loc1+1];
+    //assert(type == map_type(T));
+
+    std::string str_ws = header.substr(loc1+2);
+    loc2 = str_ws.find("'");
+    word_size = atoi(str_ws.substr(0,loc2).c_str());
+}
+
+void cnpy::parse_zip_footer(FILE* fp, unsigned short& nrecs, unsigned int& global_header_size, unsigned int& global_header_offset)
+{
+    std::vector<char> footer(22);
+    fseek(fp,-22,SEEK_END);
+    size_t res = fread(&footer[0],sizeof(char),22,fp);
+    if(res != 22)
+        throw std::runtime_error("parse_zip_footer: failed fread");
+
+    unsigned short disk_no, disk_start, nrecs_on_disk, comment_len;
+    disk_no = *(unsigned short*) &footer[4];
+    disk_start = *(unsigned short*) &footer[6];
+    nrecs_on_disk = *(unsigned short*) &footer[8];
+    nrecs = *(unsigned short*) &footer[10];
+    global_header_size = *(unsigned int*) &footer[12];
+    global_header_offset = *(unsigned int*) &footer[16];
+    comment_len = *(unsigned short*) &footer[20];
+
+    assert(disk_no == 0);
+    assert(disk_start == 0);
+    assert(nrecs_on_disk == nrecs);
+    assert(comment_len == 0);
+}
+
+cnpy::NpyArray load_the_npy_file(FILE* fp) {
+    unsigned int* shape;
+    unsigned int ndims, word_size;
+    bool fortran_order;
+    cnpy::parse_npy_header(fp,word_size,shape,ndims,fortran_order);
+    unsigned long long size = 1; //long long so no overflow when multiplying by word_size
+    for(unsigned int i = 0;i < ndims;i++) size *= shape[i];
+
+    cnpy::NpyArray arr;
+    arr.word_size = word_size;
+    arr.shape = std::vector<unsigned int>(shape,shape+ndims);
+    delete[] shape;
+    arr.data = new char[size*word_size];
+    arr.fortran_order = fortran_order;
+    size_t nread = fread(arr.data,word_size,size,fp);
+    if(nread != size)
+        throw std::runtime_error("load_the_npy_file: failed fread");
+    return arr;
+}
+
+cnpy::npz_t cnpy::npz_load(std::string fname) {
+    FILE* fp = fopen(fname.c_str(),"rb");
+
+    if(!fp) printf("npz_load: Error! Unable to open file %s!\n",fname.c_str());
+    assert(fp);
+
+    cnpy::npz_t arrays;
+
+    while(1) {
+        std::vector<char> local_header(30);
+        size_t headerres = fread(&local_header[0],sizeof(char),30,fp);
+        if(headerres != 30)
+            throw std::runtime_error("npz_load: failed fread");
+
+        //if we've reached the global header, stop reading
+        if(local_header[2] != 0x03 || local_header[3] != 0x04) break;
+
+        //read in the variable name
+        unsigned short name_len = *(unsigned short*) &local_header[26];
+        std::string varname(name_len,' ');
+        size_t vname_res = fread(&varname[0],sizeof(char),name_len,fp);
+        if(vname_res != name_len)
+            throw std::runtime_error("npz_load: failed fread");
+
+        //erase the lagging .npy
+        varname.erase(varname.end()-4,varname.end());
+
+        //read in the extra field
+        unsigned short extra_field_len = *(unsigned short*) &local_header[28];
+        if(extra_field_len > 0) {
+            std::vector<char> buff(extra_field_len);
+            size_t efield_res = fread(&buff[0],sizeof(char),extra_field_len,fp);
+            if(efield_res != extra_field_len)
+                throw std::runtime_error("npz_load: failed fread");
+        }
+
+        arrays[varname] = load_the_npy_file(fp);
+    }
+
+    fclose(fp);
+    return arrays;
+}
+
+cnpy::NpyArray cnpy::npz_load(std::string fname, std::string varname) {
+    FILE* fp = fopen(fname.c_str(),"rb");
+
+    if(!fp) {
+        printf("npz_load: Error! Unable to open file %s!\n",fname.c_str());
+        abort();
+    }
+
+    while(1) {
+        std::vector<char> local_header(30);
+        size_t header_res = fread(&local_header[0],sizeof(char),30,fp);
+        if(header_res != 30)
+            throw std::runtime_error("npz_load: failed fread");
+
+        //if we've reached the global header, stop reading
+        if(local_header[2] != 0x03 || local_header[3] != 0x04) break;
+
+        //read in the variable name
+        unsigned short name_len = *(unsigned short*) &local_header[26];
+        std::string vname(name_len,' ');
+        size_t vname_res = fread(&vname[0],sizeof(char),name_len,fp);
+        if(vname_res != name_len)
+            throw std::runtime_error("npz_load: failed fread");
+        vname.erase(vname.end()-4,vname.end()); //erase the lagging .npy
+
+        //read in the extra field
+        unsigned short extra_field_len = *(unsigned short*) &local_header[28];
+        fseek(fp,extra_field_len,SEEK_CUR); //skip past the extra field
+
+        if(vname == varname) {
+            NpyArray array = load_the_npy_file(fp);
+            fclose(fp);
+            return array;
+        }
+        else {
+            //skip past the data
+            unsigned int size = *(unsigned int*) &local_header[22];
+            fseek(fp,size,SEEK_CUR);
+        }
+    }
+
+    fclose(fp);
+    printf("npz_load: Error! Variable name %s not found in %s!\n",varname.c_str(),fname.c_str());
+    abort();
+}
+
+cnpy::NpyArray cnpy::npy_load(std::string fname) {
+
+    FILE* fp = fopen(fname.c_str(), "rb");
+
+    if(!fp) {
+        printf("npy_load: Error! Unable to open file %s!\n",fname.c_str());
+        abort();
+    }
+
+    NpyArray arr = load_the_npy_file(fp);
+
+    fclose(fp);
+    return arr;
+}

modules/dnn/test/cnpy.h

@@ -0,0 +1,247 @@
+//Copyright (C) 2011  Carl Rogers
+//Released under MIT License
+//license available in LICENSE file, or at http://www.opensource.org/licenses/mit-license.php
+
+#ifndef LIBCNPY_H_
+#define LIBCNPY_H_
+
+#include<string>
+#include<stdexcept>
+#include<sstream>
+#include<vector>
+#include<cstdio>
+#include<typeinfo>
+#include<iostream>
+#include<cassert>
+#include<map>
+#if defined(HAVE_ZLIB) && HAVE_ZLIB
+#include<zlib.h>
+#endif
+
+namespace cnpy {
+
+    struct NpyArray {
+        char* data;
+        std::vector<unsigned int> shape;
+        unsigned int word_size;
+        bool fortran_order;
+        void destruct() {delete[] data;}
+    };
+
+    struct npz_t : public std::map<std::string, NpyArray>
+    {
+        void destruct()
+        {
+            npz_t::iterator it = this->begin();
+            for(; it != this->end(); ++it) (*it).second.destruct();
+        }
+    };
+
+    char BigEndianTest();
+    char map_type(const std::type_info& t);
+    template<typename T> std::vector<char> create_npy_header(const T* data, const unsigned int* shape, const unsigned int ndims);
+    void parse_npy_header(FILE* fp,unsigned int& word_size, unsigned int*& shape, unsigned int& ndims, bool& fortran_order);
+    void parse_zip_footer(FILE* fp, unsigned short& nrecs, unsigned int& global_header_size, unsigned int& global_header_offset);
+    npz_t npz_load(std::string fname);
+    NpyArray npz_load(std::string fname, std::string varname);
+    NpyArray npy_load(std::string fname);
+
+    template<typename T> std::vector<char>& operator+=(std::vector<char>& lhs, const T rhs) {
+        //write in little endian
+        for(char byte = 0; byte < sizeof(T); byte++) {
+            char val = *((char*)&rhs+byte);
+            lhs.push_back(val);
+        }
+        return lhs;
+    }
+
+    template<> std::vector<char>& operator+=(std::vector<char>& lhs, const std::string rhs);
+    template<> std::vector<char>& operator+=(std::vector<char>& lhs, const char* rhs);
+
+
+    template<typename T> std::string tostring(T i, int = 0, char = ' ') {
+        std::stringstream s;
+        s << i;
+        return s.str();
+    }
+
+    template<typename T> void npy_save(std::string fname, const T* data, const unsigned int* shape, const unsigned int ndims, std::string mode = "w") {
+        FILE* fp = NULL;
+
+        if(mode == "a") fp = fopen(fname.c_str(),"r+b");
+
+        if(fp) {
+            //file exists. we need to append to it. read the header, modify the array size
+            unsigned int word_size, tmp_dims;
+            unsigned int* tmp_shape = 0;
+            bool fortran_order;
+            parse_npy_header(fp,word_size,tmp_shape,tmp_dims,fortran_order);
+            assert(!fortran_order);
+
+            if(word_size != sizeof(T)) {
+                std::cout<<"libnpy error: "<<fname<<" has word size "<<word_size<<" but npy_save appending data sized "<<sizeof(T)<<"\n";
+                assert( word_size == sizeof(T) );
+            }
+            if(tmp_dims != ndims) {
+                std::cout<<"libnpy error: npy_save attempting to append misdimensioned data to "<<fname<<"\n";
+                assert(tmp_dims == ndims);
+            }
+
+            for(unsigned i = 1; i < ndims; i++) {
+                if(shape[i] != tmp_shape[i]) {
+                    std::cout<<"libnpy error: npy_save attempting to append misshaped data to "<<fname<<"\n";
+                    assert(shape[i] == tmp_shape[i]);
+                }
+            }
+            tmp_shape[0] += shape[0];
+
+            fseek(fp,0,SEEK_SET);
+            std::vector<char> header = create_npy_header(data,tmp_shape,ndims);
+            fwrite(&header[0],sizeof(char),header.size(),fp);
+            fseek(fp,0,SEEK_END);
+
+            delete[] tmp_shape;
+        }
+        else {
+            fp = fopen(fname.c_str(),"wb");
+            std::vector<char> header = create_npy_header(data,shape,ndims);
+            fwrite(&header[0],sizeof(char),header.size(),fp);
+        }
+
+        unsigned int nels = 1;
+        for(unsigned i = 0;i < ndims;i++) nels *= shape[i];
+
+        fwrite(data,sizeof(T),nels,fp);
+        fclose(fp);
+    }
+
+    template<typename T> void npz_save(std::string zipname, std::string fname, const T* data, const unsigned int* shape, const unsigned int ndims, std::string mode = "w")
+    {
+        //first, append a .npy to the fname
+        fname += ".npy";
+
+        //now, on with the show
+        FILE* fp = NULL;
+        unsigned short nrecs = 0;
+        unsigned int global_header_offset = 0;
+        std::vector<char> global_header;
+
+        if(mode == "a") fp = fopen(zipname.c_str(),"r+b");
+
+        if(fp) {
+            //zip file exists. we need to add a new npy file to it.
+            //first read the footer. this gives us the offset and size of the global header
+            //then read and store the global header.
+            //below, we will write the the new data at the start of the global header then append the global header and footer below it
+            unsigned int global_header_size;
+            parse_zip_footer(fp,nrecs,global_header_size,global_header_offset);
+            fseek(fp,global_header_offset,SEEK_SET);
+            global_header.resize(global_header_size);
+            size_t res = fread(&global_header[0],sizeof(char),global_header_size,fp);
+            if(res != global_header_size){
+                throw std::runtime_error("npz_save: header read error while adding to existing zip");
+            }
+            fseek(fp,global_header_offset,SEEK_SET);
+        }
+        else {
+            fp = fopen(zipname.c_str(),"wb");
+        }
+
+        std::vector<char> npy_header = create_npy_header(data,shape,ndims);
+
+        unsigned long nels = 1;
+        for (int m=0; m<ndims; m++ ) nels *= shape[m];
+        int nbytes = nels*sizeof(T) + npy_header.size();
+
+        //get the CRC of the data to be added
+        #if defined(HAVE_ZLIB) && HAVE_ZLIB
+        unsigned int crc = crc32(0L,(unsigned char*)&npy_header[0],npy_header.size());
+        crc = crc32(crc,(unsigned char*)data,nels*sizeof(T));
+        #else
+        unsigned int crc = 0;
+        #endif
+
+        //build the local header
+        std::vector<char> local_header;
+        local_header += "PK"; //first part of sig
+        local_header += (unsigned short) 0x0403; //second part of sig
+        local_header += (unsigned short) 20; //min version to extract
+        local_header += (unsigned short) 0; //general purpose bit flag
+        local_header += (unsigned short) 0; //compression method
+        local_header += (unsigned short) 0; //file last mod time
+        local_header += (unsigned short) 0;     //file last mod date
+        local_header += (unsigned int) crc; //crc
+        local_header += (unsigned int) nbytes; //compressed size
+        local_header += (unsigned int) nbytes; //uncompressed size
+        local_header += (unsigned short) fname.size(); //fname length
+        local_header += (unsigned short) 0; //extra field length
+        local_header += fname;
+
+        //build global header
+        global_header += "PK"; //first part of sig
+        global_header += (unsigned short) 0x0201; //second part of sig
+        global_header += (unsigned short) 20; //version made by
+        global_header.insert(global_header.end(),local_header.begin()+4,local_header.begin()+30);
+        global_header += (unsigned short) 0; //file comment length
+        global_header += (unsigned short) 0; //disk number where file starts
+        global_header += (unsigned short) 0; //internal file attributes
+        global_header += (unsigned int) 0; //external file attributes
+        global_header += (unsigned int) global_header_offset; //relative offset of local file header, since it begins where the global header used to begin
+        global_header += fname;
+
+        //build footer
+        std::vector<char> footer;
+        footer += "PK"; //first part of sig
+        footer += (unsigned short) 0x0605; //second part of sig
+        footer += (unsigned short) 0; //number of this disk
+        footer += (unsigned short) 0; //disk where footer starts
+        footer += (unsigned short) (nrecs+1); //number of records on this disk
+        footer += (unsigned short) (nrecs+1); //total number of records
+        footer += (unsigned int) global_header.size(); //nbytes of global headers
+        footer += (unsigned int) (global_header_offset + nbytes + local_header.size()); //offset of start of global headers, since global header now starts after newly written array
+        footer += (unsigned short) 0; //zip file comment length
+
+        //write everything
+        fwrite(&local_header[0],sizeof(char),local_header.size(),fp);
+        fwrite(&npy_header[0],sizeof(char),npy_header.size(),fp);
+        fwrite(data,sizeof(T),nels,fp);
+        fwrite(&global_header[0],sizeof(char),global_header.size(),fp);
+        fwrite(&footer[0],sizeof(char),footer.size(),fp);
+        fclose(fp);
+    }
+
+    template<typename T> std::vector<char> create_npy_header(const T*, const unsigned int* shape, const unsigned int ndims) {
+
+        std::vector<char> dict;
+        dict += "{'descr': '";
+        dict += BigEndianTest();
+        dict += map_type(typeid(T));
+        dict += tostring(sizeof(T));
+        dict += "', 'fortran_order': False, 'shape': (";
+        dict += tostring(shape[0]);
+        for(unsigned i = 1;i < ndims;i++) {
+            dict += ", ";
+            dict += tostring(shape[i]);
+        }
+        if(ndims == 1) dict += ",";
+        dict += "), }";
+        //pad with spaces so that preamble+dict is modulo 16 bytes. preamble is 10 bytes. dict needs to end with \n
+        int remainder = 16 - (10 + dict.size()) % 16;
+        dict.insert(dict.end(),remainder,' ');
+        dict.back() = '\n';
+
+        std::vector<char> header;
+        header += (unsigned char) 0x93;
+        header += "NUMPY";
+        header += (char) 0x01; //major version of numpy format
+        header += (char) 0x00; //minor version of numpy format
+        header += (unsigned short) dict.size();
+        header.insert(header.end(),dict.begin(),dict.end());
+
+        return header;
+    }
+
+
+}
+
+#endif

modules/dnn/test/npy_blob.hpp

@@ -0,0 +1,24 @@
+#ifndef __OPENCV_DNN_TEST_NPY_BLOB_HPP__
+#define __OPENCV_DNN_TEST_NPY_BLOB_HPP__
+#include "test_precomp.hpp"
+#include "cnpy.h"
+
+inline cv::dnn::Blob blobFromNPY(const cv::String &path)
+{
+    cnpy::NpyArray npyBlob = cnpy::npy_load(path.c_str());
+    cv::dnn::BlobShape shape((int)npyBlob.shape.size(), (int*)&npyBlob.shape[0]);
+
+    cv::dnn::Blob blob;
+    blob.fill(shape, CV_32F, npyBlob.data);
+
+    npyBlob.destruct();
+    return blob;
+}
+
+inline void saveBlobToNPY(cv::dnn::Blob &blob, const cv::String &path)
+{
+    cv::Vec4i shape = blob.shape4();
+    cnpy::npy_save(path.c_str(), blob.ptr<float>(), (unsigned*)&shape[0], 4);
+}
+
+#endif

modules/dnn/test/test_alexnet.cpp

@@ -0,0 +1,40 @@
+#include "test_precomp.hpp"
+#include "npy_blob.hpp"
+
+namespace cvtest
+{
+
+using namespace std;
+using namespace testing;
+using namespace cv;
+using namespace cv::dnn;
+
+template<typename TString>
+static std::string getTestFile(TString filename)
+{
+    return (getOpenCVExtraDir() + "/dnn/") + filename;
+}
+
+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;
+    inpMats.push_back( imread(getTestFile("alexnet_0.png")) );
+    inpMats.push_back( imread(getTestFile("alexnet_1.png")) );
+    ASSERT_TRUE(!inpMats[0].empty() && !inpMats[1].empty());
+
+    net.setBlob(".data", Blob(inpMats));
+    net.forward();
+
+    Blob out = net.getBlob("prob");
+    Blob ref = blobFromNPY(getTestFile("alexnet.npy"));
+    normAssert(ref, out, "prob");
+}
+
+}

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;
@@ -20,43 +19,24 @@ static std::string getTestFile(TStr filename)
     return (getOpenCVExtraDir() + "/dnn/") + filename;
 }
 
-TEST(ReadCaffePrototxt_gtsrb, Accuracy)
+TEST(ReadCaffe_GTSRB, Accuracy)
 {
-    Ptr<Importer> importer = createCaffeImporter(getTestFile("gtsrb.prototxt"), getTestFile("gtsrb_iter_36000.caffemodel"));
     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++)
+TEST(ReadCaffe_GoogLeNet, Accuracy)
+{
+    Net net;
     {
-        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));
+        Ptr<Importer> importer = createCaffeImporter(getTestFile("bvlc_googlenet.prototxt"), "");
+        ASSERT_TRUE(importer != NULL);
+        importer->populateNet(net);
     }
 }
 
-//TEST(ReadCaffePrototxt_GoogleNet, Accuracy)
-//{
-//    Ptr<Importer> importer = createCaffeImporter(getOpenCVExtraDir() + "/dnn/googlenet_deploy.prototxt", "");
-//    Net net;
-//    importer->populateNet(net);
-//    net.forward();
-//}
-
 }

modules/dnn/test/test_common.hpp

@@ -0,0 +1,24 @@
+#ifndef __OPENCV_TEST_COMMON_HPP__
+#define __OPENCV_TEST_COMMON_HPP__
+
+inline const std::string &getOpenCVExtraDir()
+{
+    return cvtest::TS::ptr()->get_data_path();
+}
+
+inline void normAssert(cv::InputArray ref, cv::InputArray get, const char *comment = "")
+{
+    double normL1 = cvtest::norm(ref, get, cv::NORM_L1)/ ref.getMat().total();
+    EXPECT_NEAR(normL1, 0, 0.0001) << comment;
+
+    double normInf = cvtest::norm(ref, get, cv::NORM_INF);
+    EXPECT_NEAR(normInf, 0, 0.001) << comment;
+}
+
+inline void normAssert(cv::dnn::Blob &ref, cv::dnn::Blob &test, const char *comment = "")
+{
+    EXPECT_EQ(ref.shape(), test.shape());
+    normAssert(ref.getMatRef(), test.getMatRef(), comment);
+}
+
+#endif

modules/dnn/test/test_googlenet.cpp

@@ -0,0 +1,41 @@
+#include "test_precomp.hpp"
+#include "npy_blob.hpp"
+
+namespace cvtest
+{
+
+using namespace std;
+using namespace testing;
+using namespace cv;
+using namespace cv::dnn;
+
+template<typename TString>
+static std::string getTestFile(TString filename)
+{
+    return (getOpenCVExtraDir() + "/dnn/") + filename;
+}
+
+TEST(Reproducibility_GoogLeNet, Accuracy)
+{
+    Net net;
+    {
+        Ptr<Importer> importer = createCaffeImporter(getTestFile("bvlc_googlenet.prototxt"), getTestFile("bvlc_googlenet.caffemodel"));
+        ASSERT_TRUE(importer != NULL);
+        importer->populateNet(net);
+    }
+
+    std::vector<Mat> inpMats;
+    inpMats.push_back( imread(getTestFile("googlenet_0.jpg")) );
+    inpMats.push_back( imread(getTestFile("googlenet_1.jpg")) );
+    ASSERT_TRUE(!inpMats[0].empty() && !inpMats[1].empty());
+
+    Blob inp(inpMats);
+    net.setBlob(".data", inp);
+    net.forward();
+
+    Blob out = net.getBlob("prob");
+    Blob ref = blobFromNPY(getTestFile("googlenet_prob.npy"));
+    normAssert(out, ref);
+}
+
+}

modules/dnn/test/test_layers.cpp

@@ -0,0 +1,103 @@
+#include "test_precomp.hpp"
+#include <iostream>
+#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/layers/") + filename;
+}
+
+template<typename T, int n>
+bool isEqual(const cv::Vec<T, n> &l, const cv::Vec<T, n> &r)
+{
+    for (int i = 0; i < n; i++)
+    {
+        if (l[i] != r[i])
+            return false;
+    }
+    return true;
+}
+
+static void testLayer(String proto, String caffemodel = String())
+{
+    Blob inp = blobFromNPY(getTestFile("blob.npy"));
+    Blob ref = blobFromNPY(getTestFile(proto + ".caffe.npy"));
+
+    Net net;
+    {
+        Ptr<Importer> importer = createCaffeImporter(getTestFile(proto), caffemodel);
+        ASSERT_TRUE(importer != NULL);
+        importer->populateNet(net);
+    }
+
+    net.setBlob(".input", inp);
+    net.forward();
+    Blob out = net.getBlob("output");
+
+    EXPECT_EQ(ref.shape(), out.shape());
+
+    Mat &mRef = ref.getMatRef();
+    Mat &mOut = out.getMatRef();
+
+    double normL1 = cvtest::norm(mRef, mOut, NORM_L1) / ref.total();
+    EXPECT_LE(normL1, 0.0001);
+
+    double normInf = cvtest::norm(mRef, mOut, NORM_INF);
+    EXPECT_LE(normInf, 0.0001);
+}
+
+TEST(Layer_Softmax_Test, Accuracy)
+{
+     testLayer("softmax.prototxt");
+}
+
+TEST(Layer_LRN_spatial_Test, Accuracy)
+{
+     testLayer("lrn_spatial.prototxt");
+}
+
+TEST(Layer_LRN_channels_Test, Accuracy)
+{
+     testLayer("lrn_channels.prototxt");
+}
+
+TEST(Layer_Reshape_Split_Slice_Test, Accuracy)
+{
+    Net net;
+    {
+        Ptr<Importer> importer = createCaffeImporter(getTestFile("reshape_and_slice_routines.prototxt"));
+        ASSERT_TRUE(importer != NULL);
+        importer->populateNet(net);
+    }
+
+    BlobShape shape = BlobShape(Vec2i(6, 12));
+
+    Mat1f inputMat(shape[0], shape[1]);
+    RNG rng(0);
+    rng.fill(inputMat, RNG::UNIFORM, -1, 1);
+
+    Blob input(inputMat);
+    input.reshape(shape);
+    net.setBlob(".input", input);
+    net.forward();
+    Blob output = net.getBlob("output");
+
+    input.fill(shape, CV_32F, inputMat.data);
+    normAssert(input, output);
+}
+
+}

modules/dnn/test/test_precomp.hpp

@@ -16,5 +16,6 @@
 #include "opencv2/ts.hpp"
 #include <opencv2/ts/ts_perf.hpp>
 #include <opencv2/core/utility.hpp>
+#include "test_common.hpp"
 
 #endif

modules/dnn/testdata/dnn/bvlc_alexnet.prototxt

@@ -4,7 +4,6 @@ input_dim: 10
 input_dim: 3
 input_dim: 227
 input_dim: 227
-
 layer {
   name: "conv1"
   type: "Convolution"
@@ -52,7 +51,6 @@ layer {
     stride: 2
   }
 }
-
 layer {
   name: "conv2"
   type: "Convolution"
@@ -101,7 +99,6 @@ layer {
     stride: 2
   }
 }
-
 layer {
   name: "conv3"
   type: "Convolution"
@@ -127,7 +124,6 @@ layer {
   bottom: "conv3"
   top: "conv3"
 }
-
 layer {
   name: "conv4"
   type: "Convolution"
@@ -154,7 +150,6 @@ layer {
   bottom: "conv4"
   top: "conv4"
 }
-
 layer {
   name: "conv5"
   type: "Convolution"
@@ -219,7 +214,7 @@ layer {
   name: "drop6"
   type: "Dropout"
   bottom: "fc6"
-  top: "fc61"
+  top: "fc6"
   dropout_param {
     dropout_ratio: 0.5
   }
@@ -227,7 +222,7 @@ layer {
 layer {
   name: "fc7"
   type: "InnerProduct"
-  bottom: "fc61"
+  bottom: "fc6"
   top: "fc7"
   param {
     lr_mult: 1
@@ -251,7 +246,7 @@ layer {
   name: "drop7"
   type: "Dropout"
   bottom: "fc7"
-  top: "fc71"
+  top: "fc7"
   dropout_param {
     dropout_ratio: 0.5
   }
@@ -259,7 +254,7 @@ layer {
 layer {
   name: "fc8"
   type: "InnerProduct"
-  bottom: "fc71"
+  bottom: "fc7"
   top: "fc8"
   param {
     lr_mult: 1

modules/dnn/testdata/dnn/layers/lrn_channels.prototxt

@@ -0,0 +1,21 @@
+name: "test_LRN_channels"
+input: "input"
+
+input_dim: 2
+input_dim: 6
+input_dim: 75
+input_dim: 113
+
+layer {
+  type: "LRN"
+  lrn_param {
+  	norm_region: ACROSS_CHANNELS;
+    local_size: 5
+    alpha: 1.1
+    beta: 0.75
+  }
+
+  name: "output"
+  bottom: "input"
+  top: "output"
+}

modules/dnn/testdata/dnn/layers/lrn_spatial.prototxt

@@ -0,0 +1,22 @@
+name: "test_LRN_spatial"
+input: "input"
+
+input_dim: 2
+input_dim: 6
+input_dim: 75
+input_dim: 113
+
+layer {
+  type: "LRN"
+
+  lrn_param {
+  	norm_region: WITHIN_CHANNEL;
+    local_size: 5
+    alpha: 0.9
+    beta: 0.75
+  }
+
+  name: "output"
+  bottom: "input"
+  top: "output"
+}

modules/dnn/testdata/dnn/layers/reshape_and_slice_routines.prototxt

@@ -0,0 +1,77 @@
+name: "test_reshape_splice_split"
+input: "input"
+
+layer{
+    type: "Split"
+    name: "dummy_split"
+    bottom: "input"
+    top: "dummy_split_0"
+    top: "dummy_split_1"
+}
+layer{
+    type: "Slice"
+    name: "dummy_slice_0"
+    bottom: "dummy_split_0"
+    slice_param{
+        slice_point: 1
+        slice_point: 2
+    }
+    top: "dummy_slice_0_0"
+    top: "dummy_slice_0_1"
+    top: "dummy_slice_0_2"
+}
+layer{
+    type: "Slice"
+    name: "dummy_slice_1"
+    bottom: "dummy_split_1"
+    slice_param{
+        slice_point: 1
+        slice_point: 2
+    }
+    top: "dummy_slice_1_0"
+    top: "dummy_slice_1_1"
+    top: "dummy_slice_1_2"
+}
+layer{
+    type: "Sigmoid"
+    name: "alter_sliced_split"
+    bottom: "dummy_slice_1_2"
+    top: "dummy_slice_1_2"
+}
+layer{
+    type: "Concat"
+    name: "dummy_concat"
+    bottom: "dummy_slice_0_0"
+    bottom: "dummy_slice_1_1"
+    bottom: "dummy_slice_0_2"
+    top: "dummy_concat"
+}
+layer{
+    type: "Reshape"
+    name: "dummy_reshape"
+    bottom: "dummy_concat"
+    reshape_param{
+        shape{
+            dim: 0
+            dim: 1
+            dim: 1
+            dim: -1
+            dim: 1
+        }
+        axis: 1
+        num_axes: 1
+    }
+    top: "dummy_reshape"
+}
+layer{
+    type: "Flatten"
+    name: "dummy_reshape_undo"
+    bottom: "dummy_reshape"
+    top: "dummy_reshape_undo"
+}
+layer{
+    type: "Split"
+    name: "output"
+    bottom: "dummy_reshape_undo"
+    top: "output"
+}

modules/dnn/testdata/dnn/layers/softmax.prototxt

@@ -0,0 +1,15 @@
+name: "test_Softmax"
+input: "input"
+
+input_dim: 2
+input_dim: 5
+input_dim: 75
+input_dim: 113
+
+layer {
+  type: "Softmax"
+
+  name: "output"
+  bottom: "input"
+  top: "output"
+}