Commit f3237fdc authored by Alexander Alekhin's avatar Alexander Alekhin

Merge remote-tracking branch 'upstream/3.4' into merge-3.4

parents cc259e49 a8c257ce
...@@ -64,12 +64,3 @@ else() ...@@ -64,12 +64,3 @@ else()
endif() endif()
endforeach(flag_var) endforeach(flag_var)
endif() endif()
if(CMAKE_VERSION VERSION_GREATER "2.8.6")
include(ProcessorCount)
ProcessorCount(N)
if(NOT N EQUAL 0)
SET(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} /MP${N} ")
SET(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /MP${N} ")
endif()
endif()
...@@ -444,3 +444,38 @@ if(OPENCV_EXTRA_RPATH_LINK_PATH) ...@@ -444,3 +444,38 @@ if(OPENCV_EXTRA_RPATH_LINK_PATH)
message(WARNING "OPENCV_EXTRA_RPATH_LINK_PATH may not work properly because CMAKE_EXECUTABLE_RPATH_LINK_CXX_FLAG is not defined (not supported)") message(WARNING "OPENCV_EXTRA_RPATH_LINK_PATH may not work properly because CMAKE_EXECUTABLE_RPATH_LINK_CXX_FLAG is not defined (not supported)")
endif() endif()
endif() endif()
# Control MSVC /MP flag
# Input variables: OPENCV_MSVC_PARALLEL (ON,1,2,3,...) + OPENCV_SKIP_MSVC_PARALLEL
# Details:
# - https://docs.microsoft.com/en-us/cpp/build/reference/mp-build-with-multiple-processes
# - https://docs.microsoft.com/en-us/cpp/build/reference/cl-environment-variables
# - https://gitlab.kitware.com/cmake/cmake/merge_requests/1718/diffs
if(CMAKE_GENERATOR MATCHES "Visual Studio" AND CMAKE_CXX_COMPILER_ID MATCHES "MSVC|Intel")
ocv_check_environment_variables(OPENCV_SKIP_MSVC_PARALLEL)
if(OPENCV_SKIP_MSVC_PARALLEL)
# nothing
elseif(" ${CMAKE_CXX_FLAGS}" MATCHES "/MP")
# nothing, already defined in compiler flags
elseif(DEFINED ENV{CL} AND " $ENV{CL}" MATCHES "/MP")
# nothing, compiler will use CL environment variable
elseif(DEFINED ENV{_CL_} AND " $ENV{_CL_}" MATCHES "/MP")
# nothing, compiler will use _CL_ environment variable
else()
ocv_check_environment_variables(OPENCV_MSVC_PARALLEL)
set(_mp_value "ON")
if(DEFINED OPENCV_MSVC_PARALLEL)
set(_mp_value "${OPENCV_MSVC_PARALLEL}")
endif()
set(OPENCV_MSVC_PARALLEL "${_mp_value}" CACHE STRING "Control MSVC /MP flag")
if(_mp_value)
if(_mp_value GREATER 0)
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} /MP${_mp_value}")
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /MP${_mp_value}")
else()
set(CMAKE_C_FLAGS "${CMAKE_C_FLAGS} /MP")
set(CMAKE_CXX_FLAGS "${CMAKE_CXX_FLAGS} /MP")
endif()
endif()
endif()
endif()
...@@ -92,9 +92,9 @@ endif() ...@@ -92,9 +92,9 @@ endif()
if(INF_ENGINE_TARGET) if(INF_ENGINE_TARGET)
if(NOT INF_ENGINE_RELEASE) if(NOT INF_ENGINE_RELEASE)
message(WARNING "InferenceEngine version have not been set, 2019R3 will be used by default. Set INF_ENGINE_RELEASE variable if you experience build errors.") message(WARNING "InferenceEngine version has not been set, 2020.1 will be used by default. Set INF_ENGINE_RELEASE variable if you experience build errors.")
endif() endif()
set(INF_ENGINE_RELEASE "2019030000" CACHE STRING "Force IE version, should be in form YYYYAABBCC (e.g. 2018R2.0.2 -> 2018020002)") set(INF_ENGINE_RELEASE "2020010000" CACHE STRING "Force IE version, should be in form YYYYAABBCC (e.g. 2020.1.0.2 -> 2020010002)")
set_target_properties(${INF_ENGINE_TARGET} PROPERTIES set_target_properties(${INF_ENGINE_TARGET} PROPERTIES
INTERFACE_COMPILE_DEFINITIONS "HAVE_INF_ENGINE=1;INF_ENGINE_RELEASE=${INF_ENGINE_RELEASE}" INTERFACE_COMPILE_DEFINITIONS "HAVE_INF_ENGINE=1;INF_ENGINE_RELEASE=${INF_ENGINE_RELEASE}"
) )
......
...@@ -68,6 +68,8 @@ namespace cv { ...@@ -68,6 +68,8 @@ namespace cv {
namespace dnn { namespace dnn {
CV__DNN_INLINE_NS_BEGIN CV__DNN_INLINE_NS_BEGIN
static size_t DNN_NETWORK_DUMP = utils::getConfigurationParameterSizeT("OPENCV_DNN_NETWORK_DUMP", 0);
// this option is useful to run valgrind memory errors detection // this option is useful to run valgrind memory errors detection
static bool DNN_DISABLE_MEMORY_OPTIMIZATIONS = utils::getConfigurationParameterBool("OPENCV_DNN_DISABLE_MEMORY_OPTIMIZATIONS", false); static bool DNN_DISABLE_MEMORY_OPTIMIZATIONS = utils::getConfigurationParameterBool("OPENCV_DNN_DISABLE_MEMORY_OPTIMIZATIONS", false);
...@@ -92,6 +94,7 @@ using std::vector; ...@@ -92,6 +94,7 @@ using std::vector;
using std::map; using std::map;
using std::make_pair; using std::make_pair;
using std::set; using std::set;
using std::string;
//================================================================================================== //==================================================================================================
...@@ -1113,12 +1116,19 @@ static Ptr<BackendWrapper> wrapMat(int backendId, int targetId, cv::Mat& m) ...@@ -1113,12 +1116,19 @@ static Ptr<BackendWrapper> wrapMat(int backendId, int targetId, cv::Mat& m)
return Ptr<BackendWrapper>(); // TODO Error? return Ptr<BackendWrapper>(); // TODO Error?
} }
static int g_networkId = 0;
struct Net::Impl struct Net::Impl
{ {
typedef std::map<int, LayerShapes> LayersShapesMap; typedef std::map<int, LayerShapes> LayersShapesMap;
typedef std::map<int, LayerData> MapIdToLayerData; typedef std::map<int, LayerData> MapIdToLayerData;
const int networkId; // network global identifier
int networkDumpCounter; // dump counter
Impl() Impl()
: networkId(CV_XADD(&g_networkId, 1))
, networkDumpCounter(0)
{ {
//allocate fake net input layer //allocate fake net input layer
netInputLayer = Ptr<DataLayer>(new DataLayer()); netInputLayer = Ptr<DataLayer>(new DataLayer());
...@@ -1326,6 +1336,11 @@ struct Net::Impl ...@@ -1326,6 +1336,11 @@ struct Net::Impl
{ {
CV_TRACE_FUNCTION(); CV_TRACE_FUNCTION();
if (DNN_NETWORK_DUMP > 0 && networkDumpCounter == 0)
{
dumpNetworkToFile();
}
if (preferableBackend == DNN_BACKEND_DEFAULT) if (preferableBackend == DNN_BACKEND_DEFAULT)
preferableBackend = (Backend)PARAM_DNN_BACKEND_DEFAULT; preferableBackend = (Backend)PARAM_DNN_BACKEND_DEFAULT;
#ifdef HAVE_INF_ENGINE #ifdef HAVE_INF_ENGINE
...@@ -1423,6 +1438,11 @@ struct Net::Impl ...@@ -1423,6 +1438,11 @@ struct Net::Impl
netWasAllocated = true; netWasAllocated = true;
this->blobsToKeep = blobsToKeep_; this->blobsToKeep = blobsToKeep_;
if (DNN_NETWORK_DUMP > 0)
{
dumpNetworkToFile();
}
} }
} }
...@@ -3238,6 +3258,31 @@ struct Net::Impl ...@@ -3238,6 +3258,31 @@ struct Net::Impl
static static
Net createNetworkFromModelOptimizer(InferenceEngine::CNNNetwork& ieNet); Net createNetworkFromModelOptimizer(InferenceEngine::CNNNetwork& ieNet);
#endif #endif
string dump();
void dumpNetworkToFile()
{
#ifndef OPENCV_DNN_DISABLE_NETWORK_AUTO_DUMP
String dumpFileName = cv::format("ocv_dnn_net_%05d_%02d.dot", networkId, networkDumpCounter++);
try
{
string dumpStr = dump();
std::ofstream out(dumpFileName.c_str(), std::ios::out | std::ios::binary);
out << dumpStr;
}
catch (const std::exception& e)
{
std::ofstream out((dumpFileName + ".error").c_str(), std::ios::out);
out << "Exception: " << e.what() << std::endl;
}
catch (...)
{
std::ofstream out((dumpFileName + ".error").c_str(), std::ios::out);
out << "Can't dump: unknown exception" << std::endl;
}
#endif
}
}; };
Net::Net() : impl(new Net::Impl) Net::Net() : impl(new Net::Impl)
...@@ -3756,20 +3801,26 @@ int Net::getLayerId(const String &layer) ...@@ -3756,20 +3801,26 @@ int Net::getLayerId(const String &layer)
return impl->getLayerId(layer); return impl->getLayerId(layer);
} }
String parseLayerParams(const String& name, const LayerParams& lp) { static
string dumpLayerParameterSize(const string& name, const LayerParams& lp)
{
std::ostringstream out(name, std::ios::ate);
DictValue param = lp.get(name); DictValue param = lp.get(name);
std::ostringstream out; switch (param.size())
out << name << " "; {
switch (param.size()) { case 1: out << " : "; break;
case 1: out << ": "; break; case 2: out << " (HxW): "; break;
case 2: out << "(HxW): "; break; case 3: out << " (DxHxW): "; break;
case 3: out << "(DxHxW): "; break; default:
default: CV_Error(Error::StsNotImplemented, format("Unsupported %s size = %d", name.c_str(), param.size())); CV_LOG_INFO(NULL, format("DNN/dumpLayerParameterSize(): Unsupported '%s' size = %d", name.c_str(), param.size()));
out << ": ";
} }
for (size_t i = 0; i < param.size() - 1; i++) { for (size_t i = 0; i < param.size(); i++)
out << param.get<int>(i) << " x "; {
if (i > 0)
out << " x ";
out << param.get<int>(i);
} }
out << param.get<int>(param.size() - 1) << "\\l";
return out.str(); return out.str();
} }
...@@ -3777,23 +3828,33 @@ String Net::dump() ...@@ -3777,23 +3828,33 @@ String Net::dump()
{ {
CV_Assert(!empty()); CV_Assert(!empty());
if (impl->netInputLayer->inputsData.empty()) bool hasInput = !impl->netInputLayer->inputsData.empty();
CV_Error(Error::StsError, "Requested set input");
if (hasInput)
{
if (!impl->netWasAllocated) if (!impl->netWasAllocated)
impl->setUpNet(); impl->setUpNet();
}
return impl->dump();
}
string Net::Impl::dump()
{
bool hasInput = !netInputLayer->inputsData.empty();
std::ostringstream out; std::ostringstream out;
std::map<int, LayerData>& map = impl->layers; const std::map<int, LayerData>& map = layers;
int prefBackend = impl->preferableBackend;
Backend prefBackend = (Backend)preferableBackend;
std::vector<std::vector<int> > skippedLayers; std::vector<std::vector<int> > skippedLayers;
std::vector<int> skipId; std::vector<int> skipId;
std::vector<int> allLayers(map.size(), -1); std::vector<int> allLayers(map.size(), -1);
int idPrev = -1; int idPrev = -1;
Ptr<BackendNode> prevNode; Ptr<BackendNode> prevNode;
for (std::map<int, LayerData>::reverse_iterator rit = map.rbegin(); rit != map.rend(); ++rit) for (std::map<int, LayerData>::const_reverse_iterator rit = map.rbegin(); rit != map.rend(); ++rit)
{ {
std::map<int, Ptr<BackendNode> >::iterator itBackend = rit->second.backendNodes.find(prefBackend); std::map<int, Ptr<BackendNode> >::const_iterator itBackend = rit->second.backendNodes.find(prefBackend);
if (prefBackend == DNN_BACKEND_OPENCV || itBackend == rit->second.backendNodes.end() || if (prefBackend == DNN_BACKEND_OPENCV || itBackend == rit->second.backendNodes.end() ||
itBackend->second.empty()) itBackend->second.empty())
{ {
...@@ -3832,157 +3893,208 @@ String Net::dump() ...@@ -3832,157 +3893,208 @@ String Net::dump()
prevNode = itBackend->second; prevNode = itBackend->second;
} }
} }
String colors[] = {"#ffffb3", "#fccde5", "#8dd3c7", "#bebada", "#80b1d3", "#fdb462", "#ff4848"}; string colors[] = {"#ffffb3", "#fccde5", "#8dd3c7", "#bebada", "#80b1d3", "#fdb462", "#ff4848", "#b35151"};
String backend; string backend;
switch (prefBackend) { switch (prefBackend)
{
case DNN_BACKEND_DEFAULT: backend = "DEFAULT/"; break; case DNN_BACKEND_DEFAULT: backend = "DEFAULT/"; break;
case DNN_BACKEND_HALIDE: backend = "HALIDE/"; break; case DNN_BACKEND_HALIDE: backend = "HALIDE/"; break;
case DNN_BACKEND_INFERENCE_ENGINE: // fallthru case DNN_BACKEND_INFERENCE_ENGINE: // fallthru
case DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019: backend = "DLIE/"; break; case DNN_BACKEND_INFERENCE_ENGINE_NN_BUILDER_2019: backend = "DLIE/"; break;
case DNN_BACKEND_INFERENCE_ENGINE_NGRAPH: backend = "NGRAPH/"; break; case DNN_BACKEND_INFERENCE_ENGINE_NGRAPH: backend = "NGRAPH/"; break;
case DNN_BACKEND_OPENCV: backend = "OCV/"; break; case DNN_BACKEND_OPENCV: backend = "OCV/"; break;
case DNN_BACKEND_VKCOM: backend = "VULKAN/"; break;
case DNN_BACKEND_CUDA: backend = "CUDA/"; break; case DNN_BACKEND_CUDA: backend = "CUDA/"; break;
// don't use default:
} }
out << "digraph G {" << '\n'; out << "digraph G {\n";
// Add nodes // Add nodes
for (std::map<int, LayerData>::iterator it = map.begin(); it != map.end(); ++it) for (std::map<int, LayerData>::const_iterator it = map.begin(); it != map.end(); ++it)
{ {
String name = it->second.params.name; const LayerData& ld = it->second;
if (allLayers[it->first] == -1 && !name.empty()) { string name = ld.params.name;
out << " " << "\"" << name << "\"" << " [label=\""; std::vector<int> clusterIds(1, it->first);
skipId.clear(); if (allLayers[it->first] == -1 && !name.empty())
skipId.push_back(it->first); {
out << "\t\"" << name << "\" [label=\"";
} }
else if (name.empty() || it->first != skippedLayers[allLayers[it->first]][0]) else if (name.empty() || it->first != skippedLayers[allLayers[it->first]][0])
{
continue; continue;
else { // first node in cluster : it->first == skippedLayers[allLayers[it->first]][0] }
else // first node in cluster : it->first == skippedLayers[allLayers[it->first]][0]
{
int cluster = allLayers[it->first]; int cluster = allLayers[it->first];
out << " " << "\"" << "cluster_" << cluster << "\"" << " [label=\"{"; out << "\t\"" << "cluster_" << cluster << "\" [label=\"{";
skipId = skippedLayers[allLayers[it->first]]; // vertices in current cluster clusterIds = skippedLayers[allLayers[it->first]]; // vertices in current cluster
} }
for (int i = 0; i < skipId.size(); i++) for (int i = 0; i < clusterIds.size(); i++)
{ {
LayerParams& lp = map[skipId[i]].params; CV_DbgAssert(map.find(clusterIds[i]) != map.end());
const LayerParams& lp = map.find(clusterIds[i])->second.params;
if (!lp.name.empty()) { if (!lp.name.empty()) {
if (i > 0) { if (i > 0) {
out << " | "; out << " | ";
} }
out << lp.name << "\\n" << lp.type << "\\n"; out << lp.name << "\\n" << lp.type << "\\n"; // align center
if (lp.has("kernel_size")) { if (lp.has("kernel_size"))
String kernel = parseLayerParams("kernel_size", lp); {
string kernel = dumpLayerParameterSize("kernel_size", lp);
out << kernel; out << kernel;
out << "\\l"; // align left
} else if (lp.has("kernel_h") && lp.has("kernel_w")) { } else if (lp.has("kernel_h") && lp.has("kernel_w")) {
DictValue h = lp.get("kernel_h"); DictValue h = lp.get("kernel_h");
DictValue w = lp.get("kernel_w"); DictValue w = lp.get("kernel_w");
out << "kernel (HxW): " << h << " x " << w << "\\l"; out << "kernel (HxW): " << h << " x " << w;
out << "\\l"; // align left
} }
if (lp.has("stride")) { if (lp.has("stride")) {
String stride = parseLayerParams("stride", lp); string stride = dumpLayerParameterSize("stride", lp);
out << stride; out << stride;
out << "\\l"; // align left
} else if (lp.has("stride_h") && lp.has("stride_w")) { } else if (lp.has("stride_h") && lp.has("stride_w")) {
DictValue h = lp.get("stride_h"); DictValue h = lp.get("stride_h");
DictValue w = lp.get("stride_w"); DictValue w = lp.get("stride_w");
out << "stride (HxW): " << h << " x " << w << "\\l"; out << "stride (HxW): " << h << " x " << w;
out << "\\l"; // align left
} }
if (lp.has("dilation")) { if (lp.has("dilation")) {
String dilation = parseLayerParams("dilation", lp); string dilation = dumpLayerParameterSize("dilation", lp);
out << dilation; out << dilation;
out << "\\l"; // align left
} else if (lp.has("dilation_h") && lp.has("dilation_w")) { } else if (lp.has("dilation_h") && lp.has("dilation_w")) {
DictValue h = lp.get("dilation_h"); DictValue h = lp.get("dilation_h");
DictValue w = lp.get("dilation_w"); DictValue w = lp.get("dilation_w");
out << "dilation (HxW): " << h << " x " << w << "\\l"; out << "dilation (HxW): " << h << " x " << w;
out << "\\l"; // align left
} }
if (lp.has("pad")) { if (lp.has("pad")) {
DictValue pad = lp.get("pad"); DictValue pad = lp.get("pad");
out << "pad "; out << "pad ";
switch (pad.size()) { switch (pad.size())
case 1: out << ": " << pad << "\\l"; break; {
case 2: out << "(HxW): (" << pad.get<int>(0) << " x " << pad.get<int>(1) << ")" << "\\l"; break; case 1: out << ": " << pad; break;
case 4: out << "(HxW): (" << pad.get<int>(0) << ", " << pad.get<int>(2) << ") x (" << pad.get<int>(1) << ", " << pad.get<int>(3) << ")" << "\\l"; break; case 2:
case 6: out << "(DxHxW): (" << pad.get<int>(0) << ", " << pad.get<int>(3) << ") x (" << pad.get<int>(1) << ", " << pad.get<int>(4) out << "(HxW): (" << pad.get<int>(0) << " x " << pad.get<int>(1) << ")";
<< ") x (" << pad.get<int>(2) << ", " << pad.get<int>(5) << ")" << "\\l"; break; break;
case 4:
out << "(HxW): (" << pad.get<int>(0) << ", " << pad.get<int>(2)
<< ") x (" << pad.get<int>(1) << ", " << pad.get<int>(3) << ")";
break;
case 6:
out << "(DxHxW): (" << pad.get<int>(0) << ", " << pad.get<int>(3)
<< ") x (" << pad.get<int>(1) << ", " << pad.get<int>(4)
<< ") x (" << pad.get<int>(2) << ", " << pad.get<int>(5) << ")";
break;
default: CV_Error(Error::StsNotImplemented, format("Unsupported pad size = %d", pad.size())); default: CV_Error(Error::StsNotImplemented, format("Unsupported pad size = %d", pad.size()));
} }
out << "\\l"; // align left
} else if (lp.has("pad_l") && lp.has("pad_t") && lp.has("pad_r") && lp.has("pad_b")) { } else if (lp.has("pad_l") && lp.has("pad_t") && lp.has("pad_r") && lp.has("pad_b")) {
DictValue l = lp.get("pad_l"); DictValue l = lp.get("pad_l");
DictValue t = lp.get("pad_t"); DictValue t = lp.get("pad_t");
DictValue r = lp.get("pad_r"); DictValue r = lp.get("pad_r");
DictValue b = lp.get("pad_b"); DictValue b = lp.get("pad_b");
out << "pad (HxW): (" << t << ", " << b << ") x (" << l << ", " << r << ")" << "\\l"; out << "pad (HxW): (" << t << ", " << b << ") x (" << l << ", " << r << ")";
out << "\\l"; // align left
} }
else if (lp.has("pooled_w") || lp.has("pooled_h")) { else if (lp.has("pooled_w") || lp.has("pooled_h")) {
DictValue h = lp.get("pooled_h"); DictValue h = lp.get("pooled_h");
DictValue w = lp.get("pooled_w"); DictValue w = lp.get("pooled_w");
out << "pad (HxW): " << h << " x " << w << "\\l"; out << "pad pooled (HxW): " << h << " x " << w;
out << "\\l"; // align left
} }
if (lp.has("pool")) { if (lp.has("pool")) {
out << "pool: " << lp.get("pool") << "\\l"; out << "pool: " << lp.get("pool");
out << "\\l"; // align left
} }
if (lp.has("global_pooling")) { if (lp.has("global_pooling")) {
out << "global_pooling: " << lp.get("global_pooling") << "\\l"; out << "global_pooling: " << lp.get("global_pooling");
out << "\\l"; // align left
} }
if (lp.has("group")) { if (lp.has("group")) {
out << "group: " << lp.get("group") << "\\l"; out << "group: " << lp.get("group");
out << "\\l"; // align left
} }
} }
} }
if (!it->second.outputBlobs.empty()) if (!ld.outputBlobs.empty())
out << "output: " << it->second.outputBlobs[0].size << "\\l"; {
out << "output: " << ld.outputBlobs[0].size;
out << "\\l"; // align left
}
Ptr<BackendNode> layerBackend = it->second.backendNodes[prefBackend]; Ptr<BackendNode> layerBackend;
std::map<int, Ptr<BackendNode> >::const_iterator ibn = ld.backendNodes.find(prefBackend);
if (ibn != ld.backendNodes.end())
layerBackend = ibn->second;
out << (!layerBackend.empty() ? backend : "OCV/"); out << (!layerBackend.empty() ? backend : "OCV/");
int colorId = 0; int colorId = 0;
switch (it->second.layerInstance->preferableTarget) { const Target target = ld.layerInstance.empty()
case DNN_TARGET_CPU: out << "CPU\\n"; colorId = layerBackend.empty() ? 0 : 5; break; ? DNN_TARGET_CPU
case DNN_TARGET_OPENCL: out << "OCL\\n"; colorId = 1; break; : (Target)(ld.layerInstance->preferableTarget); // TODO fix preferableTarget type
case DNN_TARGET_OPENCL_FP16: out << "OCL_FP16\\n"; colorId = 2; break; switch (target)
case DNN_TARGET_MYRIAD: out << "MYRIAD\\n"; colorId = 3; break; {
case DNN_TARGET_FPGA: out << "FPGA\\n"; colorId = 4; break; case DNN_TARGET_CPU: out << "CPU"; colorId = layerBackend.empty() ? 0 : 5; break;
case DNN_TARGET_CUDA: out << "CUDA\\n"; colorId = 5; break; case DNN_TARGET_OPENCL: out << "OCL"; colorId = 1; break;
case DNN_TARGET_CUDA_FP16: out << "CUDA_FP16\\n"; colorId = 6; break; case DNN_TARGET_OPENCL_FP16: out << "OCL_FP16"; colorId = 2; break;
} case DNN_TARGET_MYRIAD: out << "MYRIAD"; colorId = 3; break;
out << ((skipId.size() == 1)? "\" " : " }\" "); case DNN_TARGET_VULKAN: out << "VULKAN"; colorId = 7; break;
case DNN_TARGET_FPGA: out << "FPGA"; colorId = 4; break;
case DNN_TARGET_CUDA: out << "CUDA"; colorId = 5; break;
case DNN_TARGET_CUDA_FP16: out << "CUDA_FP16"; colorId = 6; break;
// don't use default:
}
out << "\\n"; // align center
out << ((clusterIds.size() == 1)? "\" " : " }\" ");
out << "fillcolor=\"" << colors[colorId] << "\" "; out << "fillcolor=\"" << colors[colorId] << "\" ";
out << "style=filled "; out << "style=filled ";
out << "shape=" << ((skipId.size() == 1)? "box" : "record") << "]" << '\n'; out << "shape=" << ((clusterIds.size() == 1)? "box" : "record") << "]\n";
} }
out << '\n'; out << '\n';
// Add edges // Add edges
int inputsSize = impl->netInputLayer->outNames.size(); int inputsSize = hasInput ? netInputLayer->outNames.size() : 0;
for (std::map<int, LayerData>::iterator it = map.begin(); it != map.end(); ++it) for (std::map<int, LayerData>::const_iterator it = map.begin(); it != map.end(); ++it)
{ {
const LayerData& ld = it->second;
if (allLayers[it->first] == -1) // node if (allLayers[it->first] == -1) // node
{ {
for (int i = 0; i < it->second.consumers.size(); i++) for (int i = 0; i < ld.consumers.size(); i++)
{ {
int outId = it->second.consumers[i].lid; int outId = ld.consumers[i].lid;
if (it == map.begin() && inputsSize > 1) if (it == map.begin() && inputsSize > 1)
out << " " << "\"" << it->second.name << "_" << i << "\"" << " -> "; out << "\t\"" << ld.name << "_" << i << "\"" << " -> ";
else else
out << " " << "\"" << it->second.name << "\"" << " -> "; out << "\t\"" << ld.name << "\"" << " -> ";
if (allLayers[outId] == -1) // node if (allLayers[outId] == -1) // node
out << "\"" << map[outId].name << "\"" << '\n'; {
CV_DbgAssert(map.find(outId) != map.end());
out << "\"" << map.find(outId)->second.name << "\"\n";
}
else // cluster else // cluster
out << "\"" << "cluster_" << allLayers[outId] << "\"" << '\n'; {
out << "\"" << "cluster_" << allLayers[outId] << "\"\n";
}
} }
} }
else if (it->first == skippedLayers[allLayers[it->first]].back()) // edges from last layer in cluster else if (it->first == skippedLayers[allLayers[it->first]].back()) // edges from last layer in cluster
{ {
for (int i = 0; i < it->second.consumers.size(); i++) for (int i = 0; i < ld.consumers.size(); i++)
{
int outId = ld.consumers[i].lid;
if (allLayers[outId] == -1) // node
{ {
int outId = it->second.consumers[i].lid; CV_DbgAssert(map.find(outId) != map.end());
if (allLayers[outId] == -1) { // node out << "\t\"" << "cluster_" << allLayers[it->first] << "\"" << " -> ";
out << " " << "\"" << "cluster_" << allLayers[it->first] << "\"" << " -> "; out << "\"" << map.find(outId)->second.name << "\"\n";
out << "\"" << map[outId].name << "\"" << '\n';
} }
else if (allLayers[outId] != allLayers[it->first]) { // another cluster else if (allLayers[outId] != allLayers[it->first]) { // another cluster
out << " " << "\"" << "cluster_" << allLayers[it->first] << "\"" << " -> "; out << "\t\"" << "cluster_" << allLayers[it->first] << "\"" << " -> ";
out << "\"" << "cluster_" << allLayers[outId] << "\"" << '\n'; out << "\"" << "cluster_" << allLayers[outId] << "\"\n";
} }
} }
} }
} }
out << "}"; out << "}\n";
return out.str(); return out.str();
} }
......
...@@ -119,13 +119,18 @@ public: ...@@ -119,13 +119,18 @@ public:
} }
}; };
ElementWiseLayer(const Func &f=Func()) : run_parallel(false) { func = f; } ElementWiseLayer(const Func &f=Func()) { func = f; }
virtual bool supportBackend(int backendId) CV_OVERRIDE virtual bool supportBackend(int backendId) CV_OVERRIDE
{ {
return func.supportBackend(backendId, this->preferableTarget); return func.supportBackend(backendId, this->preferableTarget);
} }
virtual void finalize(InputArrayOfArrays, OutputArrayOfArrays) CV_OVERRIDE
{
func.finalize();
}
virtual Ptr<BackendNode> tryAttach(const Ptr<BackendNode>& node) CV_OVERRIDE virtual Ptr<BackendNode> tryAttach(const Ptr<BackendNode>& node) CV_OVERRIDE
{ {
switch (node->backendId) switch (node->backendId)
...@@ -262,7 +267,6 @@ public: ...@@ -262,7 +267,6 @@ public:
} }
Func func; Func func;
bool run_parallel;
}; };
#ifdef HAVE_OPENCL #ifdef HAVE_OPENCL
...@@ -277,7 +281,16 @@ static String oclGetTMacro(const UMat &m) ...@@ -277,7 +281,16 @@ static String oclGetTMacro(const UMat &m)
} }
#endif #endif
struct ReLUFunctor struct BaseFunctor
{
void finalize() {}
bool tryFuse(Ptr<dnn::Layer>&) { return false; }
void getScaleShift(Mat&, Mat&) const {}
};
struct ReLUFunctor : public BaseFunctor
{ {
typedef ReLULayer Layer; typedef ReLULayer Layer;
float slope; float slope;
...@@ -421,14 +434,10 @@ struct ReLUFunctor ...@@ -421,14 +434,10 @@ struct ReLUFunctor
} }
#endif // HAVE_VULKAN #endif // HAVE_VULKAN
bool tryFuse(Ptr<dnn::Layer>&) { return false; }
void getScaleShift(Mat&, Mat&) const {}
int64 getFLOPSPerElement() const { return 1; } int64 getFLOPSPerElement() const { return 1; }
}; };
struct ReLU6Functor struct ReLU6Functor : public BaseFunctor
{ {
typedef ReLU6Layer Layer; typedef ReLU6Layer Layer;
float minValue, maxValue; float minValue, maxValue;
...@@ -548,14 +557,10 @@ struct ReLU6Functor ...@@ -548,14 +557,10 @@ struct ReLU6Functor
} }
#endif // HAVE_VULKAN #endif // HAVE_VULKAN
bool tryFuse(Ptr<dnn::Layer>&) { return false; }
void getScaleShift(Mat&, Mat&) const {}
int64 getFLOPSPerElement() const { return 2; } int64 getFLOPSPerElement() const { return 2; }
}; };
struct TanHFunctor struct TanHFunctor : public BaseFunctor
{ {
typedef TanHLayer Layer; typedef TanHLayer Layer;
...@@ -644,14 +649,10 @@ struct TanHFunctor ...@@ -644,14 +649,10 @@ struct TanHFunctor
} }
#endif // HAVE_VULKAN #endif // HAVE_VULKAN
bool tryFuse(Ptr<dnn::Layer>&) { return false; }
void getScaleShift(Mat&, Mat&) const {}
int64 getFLOPSPerElement() const { return 1; } int64 getFLOPSPerElement() const { return 1; }
}; };
struct SwishFunctor struct SwishFunctor : public BaseFunctor
{ {
typedef SwishLayer Layer; typedef SwishLayer Layer;
...@@ -740,15 +741,10 @@ struct SwishFunctor ...@@ -740,15 +741,10 @@ struct SwishFunctor
} }
#endif // HAVE_VULKAN #endif // HAVE_VULKAN
bool tryFuse(Ptr<dnn::Layer>&) { return false; }
void getScaleShift(Mat&, Mat&) const {}
int64 getFLOPSPerElement() const { return 3; } int64 getFLOPSPerElement() const { return 3; }
}; };
struct MishFunctor struct MishFunctor : public BaseFunctor
{ {
typedef MishLayer Layer; typedef MishLayer Layer;
...@@ -842,15 +838,10 @@ struct MishFunctor ...@@ -842,15 +838,10 @@ struct MishFunctor
} }
#endif // HAVE_VULKAN #endif // HAVE_VULKAN
bool tryFuse(Ptr<dnn::Layer>&) { return false; }
void getScaleShift(Mat&, Mat&) const {}
int64 getFLOPSPerElement() const { return 3; } int64 getFLOPSPerElement() const { return 3; }
}; };
struct SigmoidFunctor struct SigmoidFunctor : public BaseFunctor
{ {
typedef SigmoidLayer Layer; typedef SigmoidLayer Layer;
...@@ -939,19 +930,13 @@ struct SigmoidFunctor ...@@ -939,19 +930,13 @@ struct SigmoidFunctor
} }
#endif // HAVE_VULKAN #endif // HAVE_VULKAN
bool tryFuse(Ptr<dnn::Layer>&) { return false; }
void getScaleShift(Mat&, Mat&) const {}
int64 getFLOPSPerElement() const { return 3; } int64 getFLOPSPerElement() const { return 3; }
}; };
struct ELUFunctor struct ELUFunctor : public BaseFunctor
{ {
typedef ELULayer Layer; typedef ELULayer Layer;
explicit ELUFunctor() {}
bool supportBackend(int backendId, int) bool supportBackend(int backendId, int)
{ {
return backendId == DNN_BACKEND_OPENCV || return backendId == DNN_BACKEND_OPENCV ||
...@@ -1037,14 +1022,10 @@ struct ELUFunctor ...@@ -1037,14 +1022,10 @@ struct ELUFunctor
} }
#endif // HAVE_VULKAN #endif // HAVE_VULKAN
bool tryFuse(Ptr<dnn::Layer>&) { return false; }
void getScaleShift(Mat&, Mat&) const {}
int64 getFLOPSPerElement() const { return 2; } int64 getFLOPSPerElement() const { return 2; }
}; };
struct AbsValFunctor struct AbsValFunctor : public BaseFunctor
{ {
typedef AbsLayer Layer; typedef AbsLayer Layer;
...@@ -1139,14 +1120,10 @@ struct AbsValFunctor ...@@ -1139,14 +1120,10 @@ struct AbsValFunctor
} }
#endif // HAVE_VULKAN #endif // HAVE_VULKAN
bool tryFuse(Ptr<dnn::Layer>&) { return false; }
void getScaleShift(Mat&, Mat&) const {}
int64 getFLOPSPerElement() const { return 1; } int64 getFLOPSPerElement() const { return 1; }
}; };
struct BNLLFunctor struct BNLLFunctor : public BaseFunctor
{ {
typedef BNLLLayer Layer; typedef BNLLLayer Layer;
...@@ -1236,23 +1213,19 @@ struct BNLLFunctor ...@@ -1236,23 +1213,19 @@ struct BNLLFunctor
} }
#endif // HAVE_VULKAN #endif // HAVE_VULKAN
bool tryFuse(Ptr<dnn::Layer>&) { return false; }
void getScaleShift(Mat&, Mat&) const {}
int64 getFLOPSPerElement() const { return 5; } int64 getFLOPSPerElement() const { return 5; }
}; };
struct PowerFunctor struct PowerFunctor : public BaseFunctor
{ {
typedef PowerLayer Layer; typedef PowerLayer Layer;
float power; float power, scale, shift;
float scale; float originPower, originScale, originShift;
float shift;
explicit PowerFunctor(float power_ = 1.f, float scale_ = 1.f, float shift_ = 0.f) explicit PowerFunctor(float power_ = 1.f, float scale_ = 1.f, float shift_ = 0.f)
: power(power_), scale(scale_), shift(shift_) {} : power(power_), scale(scale_), shift(shift_),
originPower(power_), originScale(scale_), originShift(shift_) {}
bool supportBackend(int backendId, int targetId) bool supportBackend(int backendId, int targetId)
{ {
...@@ -1266,6 +1239,13 @@ struct PowerFunctor ...@@ -1266,6 +1239,13 @@ struct PowerFunctor
backendId == DNN_BACKEND_HALIDE; backendId == DNN_BACKEND_HALIDE;
} }
void finalize()
{
power = originPower;
scale = originScale;
shift = originShift;
}
void apply(const float* srcptr, float* dstptr, int len, size_t planeSize, int cn0, int cn1) const void apply(const float* srcptr, float* dstptr, int len, size_t planeSize, int cn0, int cn1) const
{ {
float a = scale, b = shift, p = power; float a = scale, b = shift, p = power;
...@@ -1410,8 +1390,7 @@ struct PowerFunctor ...@@ -1410,8 +1390,7 @@ struct PowerFunctor
int64 getFLOPSPerElement() const { return power == 1 ? 2 : 10; } int64 getFLOPSPerElement() const { return power == 1 ? 2 : 10; }
}; };
struct ChannelsPReLUFunctor : public BaseFunctor
struct ChannelsPReLUFunctor
{ {
typedef ChannelsPReLULayer Layer; typedef ChannelsPReLULayer Layer;
Mat scale; Mat scale;
...@@ -1545,10 +1524,6 @@ struct ChannelsPReLUFunctor ...@@ -1545,10 +1524,6 @@ struct ChannelsPReLUFunctor
} }
#endif // HAVE_VULKAN #endif // HAVE_VULKAN
bool tryFuse(Ptr<dnn::Layer>&) { return false; }
void getScaleShift(Mat&, Mat&) const {}
int64 getFLOPSPerElement() const { return 1; } int64 getFLOPSPerElement() const { return 1; }
}; };
......
...@@ -23,10 +23,11 @@ ...@@ -23,10 +23,11 @@
#define INF_ENGINE_RELEASE_2019R1 2019010000 #define INF_ENGINE_RELEASE_2019R1 2019010000
#define INF_ENGINE_RELEASE_2019R2 2019020000 #define INF_ENGINE_RELEASE_2019R2 2019020000
#define INF_ENGINE_RELEASE_2019R3 2019030000 #define INF_ENGINE_RELEASE_2019R3 2019030000
#define INF_ENGINE_RELEASE_2020_1 2020010000
#ifndef INF_ENGINE_RELEASE #ifndef INF_ENGINE_RELEASE
#warning("IE version have not been provided via command-line. Using 2019R3 by default") #warning("IE version have not been provided via command-line. Using 2019.1 by default")
#define INF_ENGINE_RELEASE INF_ENGINE_RELEASE_2019R3 #define INF_ENGINE_RELEASE INF_ENGINE_RELEASE_2020_1
#endif #endif
#define INF_ENGINE_VER_MAJOR_GT(ver) (((INF_ENGINE_RELEASE) / 10000) > ((ver) / 10000)) #define INF_ENGINE_VER_MAJOR_GT(ver) (((INF_ENGINE_RELEASE) / 10000) > ((ver) / 10000))
...@@ -49,7 +50,7 @@ ...@@ -49,7 +50,7 @@
#pragma warning(disable: 4996) // was declared deprecated #pragma warning(disable: 4996) // was declared deprecated
#endif #endif
#if defined(__GNUC__) #if defined(__GNUC__) && INF_ENGINE_VER_MAJOR_LT(INF_ENGINE_RELEASE_2020_1)
#pragma GCC visibility push(default) #pragma GCC visibility push(default)
#endif #endif
...@@ -57,7 +58,7 @@ ...@@ -57,7 +58,7 @@
#include <ie_builders.hpp> #include <ie_builders.hpp>
#if defined(__GNUC__) #if defined(__GNUC__) && INF_ENGINE_VER_MAJOR_LT(INF_ENGINE_RELEASE_2020_1)
#pragma GCC visibility pop #pragma GCC visibility pop
#endif #endif
......
...@@ -1297,7 +1297,7 @@ protected: ...@@ -1297,7 +1297,7 @@ protected:
{ {
public: public:
ParallelSearch(vector< vector< Point2f > >& true_points_group_, ParallelSearch(vector< vector< Point2f > >& true_points_group_,
vector< vector< Point2f > >& loc_, int iter_, int* end_, vector< vector< Point2f > >& loc_, int iter_, vector<int>& end_,
vector< vector< Vec3i > >& all_points_, vector< vector< Vec3i > >& all_points_,
QRDetectMulti& cl_) QRDetectMulti& cl_)
: :
...@@ -1313,7 +1313,7 @@ protected: ...@@ -1313,7 +1313,7 @@ protected:
vector< vector< Point2f > >& true_points_group; vector< vector< Point2f > >& true_points_group;
vector< vector< Point2f > >& loc; vector< vector< Point2f > >& loc;
int iter; int iter;
int* end; vector<int>& end;
vector< vector< Vec3i > >& all_points; vector< vector< Vec3i > >& all_points;
QRDetectMulti& cl; QRDetectMulti& cl;
}; };
...@@ -1925,7 +1925,7 @@ bool QRDetectMulti::checkSets(vector<vector<Point2f> >& true_points_group, vecto ...@@ -1925,7 +1925,7 @@ bool QRDetectMulti::checkSets(vector<vector<Point2f> >& true_points_group, vecto
return false; return false;
int* set_size = new int[true_points_group.size()]; vector<int> set_size(true_points_group.size());
for (size_t i = 0; i < true_points_group.size(); i++) for (size_t i = 0; i < true_points_group.size(); i++)
{ {
set_size[i] = int(0.5 * (true_points_group[i].size() - 2 ) * (true_points_group[i].size() - 1)); set_size[i] = int(0.5 * (true_points_group[i].size() - 2 ) * (true_points_group[i].size() - 1));
...@@ -1963,7 +1963,7 @@ bool QRDetectMulti::checkSets(vector<vector<Point2f> >& true_points_group, vecto ...@@ -1963,7 +1963,7 @@ bool QRDetectMulti::checkSets(vector<vector<Point2f> >& true_points_group, vecto
transformation_points.resize(iter + true_points_group.size()); transformation_points.resize(iter + true_points_group.size());
true_points_group_copy = true_points_group; true_points_group_copy = true_points_group;
int* end = new int[true_points_group.size()]; vector<int> end(true_points_group.size());
for (size_t i = 0; i < true_points_group.size(); i++) for (size_t i = 0; i < true_points_group.size(); i++)
end[i] = iter + set_size[i]; end[i] = iter + set_size[i];
ParallelSearch parallelSearch(true_points_group, ParallelSearch parallelSearch(true_points_group,
......
Markdown is supported
0% or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment