Skip to content

Commit

Permalink
Added function to get timings for layers
Browse files Browse the repository at this point in the history
  • Loading branch information
@arrybn
arrybn committed Aug 23, 2017
1 parent da0a36c commit 8b1146d
Show file tree
Hide file tree
Showing 5 changed files with 98 additions and 52 deletions.
34 changes: 22 additions & 12 deletions modules/dnn/include/opencv2/dnn/dnn.hpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -483,15 +483,15 @@ CV__DNN_EXPERIMENTAL_NS_BEGIN
* order is the same as in layersIds
*/
CV_WRAP void getLayersShapes(const std::vector<MatShape>& netInputShapes,
std::vector<int>* layersIds,
std::vector<std::vector<MatShape> >* inLayersShapes,
std::vector<std::vector<MatShape> >* outLayersShapes) const;
CV_OUT std::vector<int>& layersIds,
CV_OUT std::vector<std::vector<MatShape> >& inLayersShapes,
CV_OUT std::vector<std::vector<MatShape> >& outLayersShapes) const;

/** @overload */
CV_WRAP void getLayersShapes(const MatShape& netInputShape,
std::vector<int>* layersIds,
std::vector<std::vector<MatShape> >* inLayersShapes,
std::vector<std::vector<MatShape> >* outLayersShapes) const;
CV_OUT std::vector<int>& layersIds,
CV_OUT std::vector<std::vector<MatShape> >& inLayersShapes,
CV_OUT std::vector<std::vector<MatShape> >& outLayersShapes) const;

/** @brief Returns input and output shapes for layer with specified
* id in loaded model; preliminary inferencing isn't necessary.
Expand All @@ -504,14 +504,14 @@ CV__DNN_EXPERIMENTAL_NS_BEGIN
*/
CV_WRAP void getLayerShapes(const MatShape& netInputShape,
const int layerId,
std::vector<MatShape>* inLayerShapes,
std::vector<MatShape>* outLayerShapes) const;
CV_OUT std::vector<MatShape>& inLayerShapes,
CV_OUT std::vector<MatShape>& outLayerShapes) const;

/** @overload */
CV_WRAP void getLayerShapes(const std::vector<MatShape>& netInputShapes,
const int layerId,
std::vector<MatShape>* inLayerShapes,
std::vector<MatShape>* outLayerShapes) const;
CV_OUT std::vector<MatShape>& inLayerShapes,
CV_OUT std::vector<MatShape>& outLayerShapes) const;

/** @brief Computes FLOP for whole loaded model with specified input shapes.
* @param netInputShapes vector of shapes for all net inputs.
Expand Down Expand Up @@ -566,18 +566,28 @@ CV__DNN_EXPERIMENTAL_NS_BEGIN
* @param blobs output parameter to store resulting bytes for intermediate blobs.
*/
CV_WRAP void getMemoryConsumption(const std::vector<MatShape>& netInputShapes,
CV_OUT std::vector<int>& layerIds, CV_OUT std::vector<size_t>& weights,
CV_OUT std::vector<int>& layerIds,
CV_OUT std::vector<size_t>& weights,
CV_OUT std::vector<size_t>& blobs) const;
/** @overload */
CV_WRAP void getMemoryConsumption(const MatShape& netInputShape,
CV_OUT std::vector<int>& layerIds, CV_OUT std::vector<size_t>& weights,
CV_OUT std::vector<int>& layerIds,
CV_OUT std::vector<size_t>& weights,
CV_OUT std::vector<size_t>& blobs) const;

/** @brief Enables or disables layer fusion in the network.
* @param fusion true to enable the fusion, false to disable. The fusion is enabled by default.
*/
CV_WRAP void enableFusion(bool fusion);

/** @brief Returns overall time for inference and timings (in ticks) for layers.
* Indexes in returned vector correspond to layers ids. Some layers can be fused with others,
* in this case zero ticks count will be return for that skipped layers.
* @param timings vector for tick timings for all layers.
* @return overall ticks for model inference.
*/
CV_WRAP int64 getPerfProfile(CV_OUT std::vector<double>& timings);

private:
struct Impl;
Ptr<Impl> impl;
Expand Down
77 changes: 45 additions & 32 deletions modules/dnn/src/dnn.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -47,6 +47,7 @@
#include <iostream>
#include <sstream>
#include <iterator>
#include <numeric>
#include <opencv2/dnn/shape_utils.hpp>
#include <opencv2/imgproc.hpp>

Expand Down Expand Up @@ -633,7 +634,7 @@ struct Net::Impl
inpl.layerInstance = netInputLayer;
layerNameToId.insert(std::make_pair(inpl.name, inpl.id));

lastLayerId = 1;
lastLayerId = 0;
netWasAllocated = false;
fusion = true;
preferableBackend = DNN_BACKEND_DEFAULT;
Expand All @@ -656,6 +657,7 @@ struct Net::Impl

bool netWasAllocated;
bool fusion;
std::vector<int64> layersTimings;

void compileHalide()
{
Expand Down Expand Up @@ -716,6 +718,8 @@ struct Net::Impl
it = layers.find(0);
CV_Assert(it != layers.end());
it->second.skipFlags[DNN_BACKEND_DEFAULT] = true;

layersTimings.clear();
}

void setUpNet(const std::vector<LayerPin>& blobsToKeep_ = std::vector<LayerPin>())
Expand Down Expand Up @@ -1269,6 +1273,8 @@ struct Net::Impl
allocateLayer(lid, layersShapes);
}

layersTimings.resize(lastLayerId + 1, 0);

fuseLayers(blobsToKeep_);
}

Expand All @@ -1278,11 +1284,16 @@ struct Net::Impl

Ptr<Layer> layer = ld.layerInstance;

TickMeter tm;
tm.start();

if (preferableBackend == DNN_BACKEND_DEFAULT ||
!layer->supportBackend(preferableBackend))
{
if( !ld.skipFlags[DNN_BACKEND_DEFAULT] )
layer->forward(ld.inputBlobs, ld.outputBlobs, ld.internals);
else
tm.reset();
}
else if (!ld.skipFlags[preferableBackend])
{
Expand All @@ -1299,6 +1310,9 @@ struct Net::Impl
}
}

tm.stop();
layersTimings[ld.id] = tm.getTimeTicks();

ld.flag = 1;
}

Expand Down Expand Up @@ -1717,45 +1731,39 @@ std::vector<int> Net::getUnconnectedOutLayers() const
}

void Net::getLayersShapes(const ShapesVec& netInputShapes,
std::vector<int>* layersIds,
std::vector<ShapesVec>* inLayersShapes,
std::vector<ShapesVec>* outLayersShapes) const
std::vector<int>& layersIds,
std::vector<ShapesVec>& inLayersShapes,
std::vector<ShapesVec>& outLayersShapes) const
{
if ((layersIds || inLayersShapes || outLayersShapes) == false)
return;

if (layersIds) layersIds->clear();
if (inLayersShapes) inLayersShapes->clear();
if (outLayersShapes) outLayersShapes->clear();
layersIds.clear();
inLayersShapes.clear();
outLayersShapes.clear();

Impl::LayersShapesMap inOutShapes;
impl->getLayersShapes(netInputShapes, inOutShapes);

for(Impl::LayersShapesMap::const_iterator it = inOutShapes.begin();
it != inOutShapes.end(); it++)
{
if (layersIds)
layersIds->push_back(it->first);
if (inLayersShapes)
inLayersShapes->push_back(it->second.in);
if (outLayersShapes)
outLayersShapes->push_back(it->second.out);
layersIds.push_back(it->first);
inLayersShapes.push_back(it->second.in);
outLayersShapes.push_back(it->second.out);
}
}

void Net::getLayersShapes(const MatShape& netInputShape,
std::vector<int>* layerIds,
std::vector<ShapesVec>* inLayersShapes,
std::vector<ShapesVec>* outLayersShapes) const
std::vector<int>& layerIds,
std::vector<ShapesVec>& inLayersShapes,
std::vector<ShapesVec>& outLayersShapes) const
{
getLayersShapes(ShapesVec(1, netInputShape),
layerIds, inLayersShapes, outLayersShapes);
}

void Net::getLayerShapes(const MatShape& netInputShape,
const int layerId,
ShapesVec* inLayerShapes,
ShapesVec* outLayerShapes) const
ShapesVec& inLayerShapes,
ShapesVec& outLayerShapes) const
{
getLayerShapes(ShapesVec(1, netInputShape),
layerId, inLayerShapes, outLayerShapes);
Expand All @@ -1764,15 +1772,13 @@ void Net::getLayerShapes(const MatShape& netInputShape,

void Net::getLayerShapes(const ShapesVec& netInputShapes,
const int layerId,
ShapesVec* inLayerShapes,
ShapesVec* outLayerShapes) const
ShapesVec& inLayerShapes,
ShapesVec& outLayerShapes) const
{
LayerShapes shapes;
impl->getLayerShapes(netInputShapes, layerId, shapes);
if (inLayerShapes)
*inLayerShapes = shapes.in;
if (outLayerShapes)
*outLayerShapes = shapes.out;
inLayerShapes = shapes.in;
outLayerShapes = shapes.out;
}

int64 Net::getFLOPS(const std::vector<MatShape>& netInputShapes) const
Expand All @@ -1782,7 +1788,7 @@ int64 Net::getFLOPS(const std::vector<MatShape>& netInputShapes) const
int64 flops = 0;
std::vector<int> ids;
std::vector<std::vector<MatShape> > inShapes, outShapes;
getLayersShapes(netInputShapes, &ids, &inShapes, &outShapes);
getLayersShapes(netInputShapes, ids, inShapes, outShapes);
CV_Assert(inShapes.size() == outShapes.size());
CV_Assert(inShapes.size() == ids.size());

Expand Down Expand Up @@ -1867,8 +1873,8 @@ void Net::getMemoryConsumption(const int layerId,
weights += weightsBlob.total()*weightsBlob.elemSize();
}

std::vector<MatShape> outLayerShapes;
getLayerShapes(netInputShapes, layerId, 0, &outLayerShapes);
ShapesVec inLayerShapes, outLayerShapes;
getLayerShapes(netInputShapes, layerId, inLayerShapes, outLayerShapes);
for(int i = 0; i < outLayerShapes.size(); i++)
{
blobs += total(outLayerShapes[i]) * sizeof(float);
Expand Down Expand Up @@ -1917,9 +1923,9 @@ void Net::getMemoryConsumption(const std::vector<MatShape>& netInputShapes,
weights.clear();
blobs.clear();

std::vector<std::vector<MatShape> > outLayerShapes;
std::vector<std::vector<MatShape> > inLayerShapes, outLayerShapes;

getLayersShapes(netInputShapes, &layerIds, 0, &outLayerShapes);
getLayersShapes(netInputShapes, layerIds, inLayerShapes, outLayerShapes);

for(int i = 0; i < layerIds.size(); i++)
{
Expand Down Expand Up @@ -1968,6 +1974,13 @@ void Net::setHalideScheduler(const String& scheduler)
impl->halideConfigFile = scheduler;
}

int64 Net::getPerfProfile(std::vector<double>& timings)
{
timings = std::vector<double>(impl->layersTimings.begin() + 1, impl->layersTimings.end());
int64 total = std::accumulate(timings.begin(), timings.end(), 0);
return total;
}

//////////////////////////////////////////////////////////////////////////

Importer::~Importer() {}
Expand Down
19 changes: 18 additions & 1 deletion modules/java/generator/src/cpp/listconverters.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -56,4 +56,21 @@ void Copy_vector_String_to_List(JNIEnv* env, std::vector<cv::String>& vs, jobjec
env->CallBooleanMethod(list, m_add, element);
env->DeleteLocalRef(element);
}
}
}

#if defined(HAVE_OPENCV_DNN)
void Copy_vector_MatShape_to_List(JNIEnv* env, std::vector<cv::dnn::MatShape>& vs, jobject list)
{
static jclass juArrayList = ARRAYLIST(env);
jmethodID m_clear = LIST_CLEAR(env, juArrayList);
jmethodID m_add = LIST_ADD(env, juArrayList);

env->CallVoidMethod(list, m_clear);
for (std::vector<cv::dnn::MatShape>::iterator it = vs.begin(); it != vs.end(); ++it)
{
jstring element = env->NewStringUTF("");
env->CallBooleanMethod(list, m_add, element);
env->DeleteLocalRef(element);
}
}
#endif
7 changes: 6 additions & 1 deletion modules/java/generator/src/cpp/listconverters.hpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -16,4 +16,9 @@ std::vector<cv::String> List_to_vector_String(JNIEnv* env, jobject list);

void Copy_vector_String_to_List(JNIEnv* env, std::vector<cv::String>& vs, jobject list);

#endif /* LISTCONVERTERS_HPP */
#if defined(HAVE_OPENCV_DNN)
#include "opencv2/dnn.hpp"
void Copy_vector_MatShape_to_List(JNIEnv* env, std::vector<cv::dnn::MatShape>& vs, jobject list);
#endif

#endif /* LISTCONVERTERS_HPP */
13 changes: 7 additions & 6 deletions samples/dnn/ssd_mobilenet_object_detection.cpp
View file
Original file line number Diff line number Diff line change
Expand Up @@ -101,17 +101,18 @@ int main(int argc, char** argv)
//! [Prepare blob]

//! [Set input blob]
net.setInput(inputBlob, "data"); //set the network input
net.setInput(inputBlob, "data"); //set the network input
//! [Set input blob]

TickMeter tm;
tm.start();
//! [Make forward pass]
Mat detection = net.forward("detection_out"); //compute output
tm.stop();
cout << "Inference time, ms: " << tm.getTimeMilli() << endl;
Mat detection = net.forward("detection_out"); //compute output
//! [Make forward pass]

std::vector<double> layersTimings;
double freq = getTickFrequency() / 1000;
double time = net.getPerfProfile(layersTimings) / freq;
cout << "Inference time, ms: " << time << endl;

Mat detectionMat(detection.size[2], detection.size[3], CV_32F, detection.ptr<float>());

frame = frame(crop);
Expand Down

0 comments on commit 8b1146d

Please sign in to comment.