Commit f96f9344 authored by Dmitry Kurtaev's avatar Dmitry Kurtaev Committed by Vadim Pisarevsky

Update Intel's Inference Engine deep learning backend (#11587)

* Update Intel's Inference Engine deep learning backend

* Remove cpu_extension dependency

* Update Darknet accuracy tests
parent 80770aac
...@@ -41,8 +41,7 @@ set(INF_ENGINE_INCLUDE_DIRS "${INF_ENGINE_ROOT_DIR}/include" CACHE PATH "Path to ...@@ -41,8 +41,7 @@ set(INF_ENGINE_INCLUDE_DIRS "${INF_ENGINE_ROOT_DIR}/include" CACHE PATH "Path to
if(NOT INF_ENGINE_ROOT_DIR if(NOT INF_ENGINE_ROOT_DIR
OR NOT EXISTS "${INF_ENGINE_ROOT_DIR}" OR NOT EXISTS "${INF_ENGINE_ROOT_DIR}"
OR NOT EXISTS "${INF_ENGINE_INCLUDE_DIRS}" OR NOT EXISTS "${INF_ENGINE_ROOT_DIR}/include/inference_engine.hpp"
OR NOT EXISTS "${INF_ENGINE_INCLUDE_DIRS}/inference_engine.hpp"
) )
ie_fail() ie_fail()
endif() endif()
...@@ -52,10 +51,7 @@ set(INF_ENGINE_LIBRARIES "") ...@@ -52,10 +51,7 @@ set(INF_ENGINE_LIBRARIES "")
set(ie_lib_list inference_engine) set(ie_lib_list inference_engine)
link_directories( link_directories(
${INTEL_CVSDK_DIR}/external/mklml_lnx/lib
${INTEL_CVSDK_DIR}/inference_engine/external/mklml_lnx/lib
${INTEL_CVSDK_DIR}/inference_engine/external/mkltiny_lnx/lib ${INTEL_CVSDK_DIR}/inference_engine/external/mkltiny_lnx/lib
${INTEL_CVSDK_DIR}/external/cldnn/lib
${INTEL_CVSDK_DIR}/inference_engine/external/cldnn/lib ${INTEL_CVSDK_DIR}/inference_engine/external/cldnn/lib
) )
......
...@@ -81,7 +81,8 @@ CV__DNN_EXPERIMENTAL_NS_BEGIN ...@@ -81,7 +81,8 @@ CV__DNN_EXPERIMENTAL_NS_BEGIN
{ {
DNN_TARGET_CPU, DNN_TARGET_CPU,
DNN_TARGET_OPENCL, DNN_TARGET_OPENCL,
DNN_TARGET_OPENCL_FP16 DNN_TARGET_OPENCL_FP16,
DNN_TARGET_MYRIAD
}; };
/** @brief This class provides all data needed to initialize layer. /** @brief This class provides all data needed to initialize layer.
...@@ -700,13 +701,13 @@ CV__DNN_EXPERIMENTAL_NS_BEGIN ...@@ -700,13 +701,13 @@ CV__DNN_EXPERIMENTAL_NS_BEGIN
* * `*.pb` (TensorFlow, https://www.tensorflow.org/) * * `*.pb` (TensorFlow, https://www.tensorflow.org/)
* * `*.t7` | `*.net` (Torch, http://torch.ch/) * * `*.t7` | `*.net` (Torch, http://torch.ch/)
* * `*.weights` (Darknet, https://pjreddie.com/darknet/) * * `*.weights` (Darknet, https://pjreddie.com/darknet/)
* * `*.bin` (DLDT, https://software.seek.intel.com/deep-learning-deployment) * * `*.bin` (DLDT, https://software.intel.com/openvino-toolkit)
* @param[in] config Text file contains network configuration. It could be a * @param[in] config Text file contains network configuration. It could be a
* file with the following extensions: * file with the following extensions:
* * `*.prototxt` (Caffe, http://caffe.berkeleyvision.org/) * * `*.prototxt` (Caffe, http://caffe.berkeleyvision.org/)
* * `*.pbtxt` (TensorFlow, https://www.tensorflow.org/) * * `*.pbtxt` (TensorFlow, https://www.tensorflow.org/)
* * `*.cfg` (Darknet, https://pjreddie.com/darknet/) * * `*.cfg` (Darknet, https://pjreddie.com/darknet/)
* * `*.xml` (DLDT, https://software.seek.intel.com/deep-learning-deployment) * * `*.xml` (DLDT, https://software.intel.com/openvino-toolkit)
* @param[in] framework Explicit framework name tag to determine a format. * @param[in] framework Explicit framework name tag to determine a format.
* @returns Net object. * @returns Net object.
* *
......
...@@ -13,7 +13,7 @@ ...@@ -13,7 +13,7 @@
namespace opencv_test { namespace opencv_test {
CV_ENUM(DNNBackend, DNN_BACKEND_DEFAULT, DNN_BACKEND_HALIDE, DNN_BACKEND_INFERENCE_ENGINE) CV_ENUM(DNNBackend, DNN_BACKEND_DEFAULT, DNN_BACKEND_HALIDE, DNN_BACKEND_INFERENCE_ENGINE)
CV_ENUM(DNNTarget, DNN_TARGET_CPU, DNN_TARGET_OPENCL, DNN_TARGET_OPENCL_FP16) CV_ENUM(DNNTarget, DNN_TARGET_CPU, DNN_TARGET_OPENCL, DNN_TARGET_OPENCL_FP16, DNN_TARGET_MYRIAD)
class DNNTestNetwork : public ::perf::TestBaseWithParam< tuple<DNNBackend, DNNTarget> > class DNNTestNetwork : public ::perf::TestBaseWithParam< tuple<DNNBackend, DNNTarget> >
{ {
...@@ -29,6 +29,28 @@ public: ...@@ -29,6 +29,28 @@ public:
target = (dnn::Target)(int)get<1>(GetParam()); target = (dnn::Target)(int)get<1>(GetParam());
} }
static bool checkMyriadTarget()
{
#ifndef HAVE_INF_ENGINE
return false;
#endif
cv::dnn::Net net;
cv::dnn::LayerParams lp;
net.addLayerToPrev("testLayer", "Identity", lp);
net.setPreferableBackend(cv::dnn::DNN_BACKEND_INFERENCE_ENGINE);
net.setPreferableTarget(cv::dnn::DNN_TARGET_MYRIAD);
net.setInput(cv::Mat::zeros(1, 1, CV_32FC1));
try
{
net.forward();
}
catch(...)
{
return false;
}
return true;
}
void processNet(std::string weights, std::string proto, std::string halide_scheduler, void processNet(std::string weights, std::string proto, std::string halide_scheduler,
const Mat& input, const std::string& outputLayer = "") const Mat& input, const std::string& outputLayer = "")
{ {
...@@ -41,6 +63,13 @@ public: ...@@ -41,6 +63,13 @@ public:
throw cvtest::SkipTestException("OpenCL is not available/disabled in OpenCV"); throw cvtest::SkipTestException("OpenCL is not available/disabled in OpenCV");
} }
} }
if (backend == DNN_BACKEND_INFERENCE_ENGINE && target == DNN_TARGET_MYRIAD)
{
if (!checkMyriadTarget())
{
throw SkipTestException("Myriad is not available/disabled in OpenCV");
}
}
randu(input, 0.0f, 1.0f); randu(input, 0.0f, 1.0f);
...@@ -87,8 +116,6 @@ public: ...@@ -87,8 +116,6 @@ public:
PERF_TEST_P_(DNNTestNetwork, AlexNet) PERF_TEST_P_(DNNTestNetwork, AlexNet)
{ {
if (backend == DNN_BACKEND_INFERENCE_ENGINE && target != DNN_TARGET_CPU)
throw SkipTestException("");
processNet("dnn/bvlc_alexnet.caffemodel", "dnn/bvlc_alexnet.prototxt", processNet("dnn/bvlc_alexnet.caffemodel", "dnn/bvlc_alexnet.prototxt",
"alexnet.yml", Mat(cv::Size(227, 227), CV_32FC3)); "alexnet.yml", Mat(cv::Size(227, 227), CV_32FC3));
} }
...@@ -130,7 +157,6 @@ PERF_TEST_P_(DNNTestNetwork, ENet) ...@@ -130,7 +157,6 @@ PERF_TEST_P_(DNNTestNetwork, ENet)
PERF_TEST_P_(DNNTestNetwork, SSD) PERF_TEST_P_(DNNTestNetwork, SSD)
{ {
if (backend == DNN_BACKEND_INFERENCE_ENGINE) throw SkipTestException("");
processNet("dnn/VGG_ILSVRC2016_SSD_300x300_iter_440000.caffemodel", "dnn/ssd_vgg16.prototxt", "disabled", processNet("dnn/VGG_ILSVRC2016_SSD_300x300_iter_440000.caffemodel", "dnn/ssd_vgg16.prototxt", "disabled",
Mat(cv::Size(300, 300), CV_32FC3)); Mat(cv::Size(300, 300), CV_32FC3));
} }
...@@ -146,18 +172,17 @@ PERF_TEST_P_(DNNTestNetwork, OpenFace) ...@@ -146,18 +172,17 @@ PERF_TEST_P_(DNNTestNetwork, OpenFace)
PERF_TEST_P_(DNNTestNetwork, MobileNet_SSD_Caffe) PERF_TEST_P_(DNNTestNetwork, MobileNet_SSD_Caffe)
{ {
if (backend == DNN_BACKEND_HALIDE || if (backend == DNN_BACKEND_HALIDE)
backend == DNN_BACKEND_INFERENCE_ENGINE && target != DNN_TARGET_CPU)
throw SkipTestException(""); throw SkipTestException("");
processNet("dnn/MobileNetSSD_deploy.caffemodel", "dnn/MobileNetSSD_deploy.prototxt", "", processNet("dnn/MobileNetSSD_deploy.caffemodel", "dnn/MobileNetSSD_deploy.prototxt", "",
Mat(cv::Size(300, 300), CV_32FC3)); Mat(cv::Size(300, 300), CV_32FC3));
} }
// TODO: update MobileNet model.
PERF_TEST_P_(DNNTestNetwork, MobileNet_SSD_TensorFlow) PERF_TEST_P_(DNNTestNetwork, MobileNet_SSD_TensorFlow)
{ {
if (backend == DNN_BACKEND_DEFAULT && target == DNN_TARGET_OPENCL || if (backend == DNN_BACKEND_HALIDE ||
backend == DNN_BACKEND_HALIDE || backend == DNN_BACKEND_INFERENCE_ENGINE)
backend == DNN_BACKEND_INFERENCE_ENGINE && target != DNN_TARGET_CPU)
throw SkipTestException(""); throw SkipTestException("");
processNet("dnn/ssd_mobilenet_v1_coco.pb", "ssd_mobilenet_v1_coco.pbtxt", "", processNet("dnn/ssd_mobilenet_v1_coco.pb", "ssd_mobilenet_v1_coco.pbtxt", "",
Mat(cv::Size(300, 300), CV_32FC3)); Mat(cv::Size(300, 300), CV_32FC3));
...@@ -166,7 +191,8 @@ PERF_TEST_P_(DNNTestNetwork, MobileNet_SSD_TensorFlow) ...@@ -166,7 +191,8 @@ PERF_TEST_P_(DNNTestNetwork, MobileNet_SSD_TensorFlow)
PERF_TEST_P_(DNNTestNetwork, DenseNet_121) PERF_TEST_P_(DNNTestNetwork, DenseNet_121)
{ {
if (backend == DNN_BACKEND_HALIDE || if (backend == DNN_BACKEND_HALIDE ||
backend == DNN_BACKEND_INFERENCE_ENGINE && target == DNN_TARGET_OPENCL_FP16) backend == DNN_BACKEND_INFERENCE_ENGINE && (target == DNN_TARGET_OPENCL_FP16 ||
target == DNN_TARGET_MYRIAD))
throw SkipTestException(""); throw SkipTestException("");
processNet("dnn/DenseNet_121.caffemodel", "dnn/DenseNet_121.prototxt", "", processNet("dnn/DenseNet_121.caffemodel", "dnn/DenseNet_121.prototxt", "",
Mat(cv::Size(224, 224), CV_32FC3)); Mat(cv::Size(224, 224), CV_32FC3));
...@@ -174,21 +200,27 @@ PERF_TEST_P_(DNNTestNetwork, DenseNet_121) ...@@ -174,21 +200,27 @@ PERF_TEST_P_(DNNTestNetwork, DenseNet_121)
PERF_TEST_P_(DNNTestNetwork, OpenPose_pose_coco) PERF_TEST_P_(DNNTestNetwork, OpenPose_pose_coco)
{ {
if (backend == DNN_BACKEND_HALIDE) throw SkipTestException(""); if (backend == DNN_BACKEND_HALIDE ||
backend == DNN_BACKEND_INFERENCE_ENGINE && target == DNN_TARGET_MYRIAD)
throw SkipTestException("");
processNet("dnn/openpose_pose_coco.caffemodel", "dnn/openpose_pose_coco.prototxt", "", processNet("dnn/openpose_pose_coco.caffemodel", "dnn/openpose_pose_coco.prototxt", "",
Mat(cv::Size(368, 368), CV_32FC3)); Mat(cv::Size(368, 368), CV_32FC3));
} }
PERF_TEST_P_(DNNTestNetwork, OpenPose_pose_mpi) PERF_TEST_P_(DNNTestNetwork, OpenPose_pose_mpi)
{ {
if (backend == DNN_BACKEND_HALIDE) throw SkipTestException(""); if (backend == DNN_BACKEND_HALIDE ||
backend == DNN_BACKEND_INFERENCE_ENGINE && target == DNN_TARGET_MYRIAD)
throw SkipTestException("");
processNet("dnn/openpose_pose_mpi.caffemodel", "dnn/openpose_pose_mpi.prototxt", "", processNet("dnn/openpose_pose_mpi.caffemodel", "dnn/openpose_pose_mpi.prototxt", "",
Mat(cv::Size(368, 368), CV_32FC3)); Mat(cv::Size(368, 368), CV_32FC3));
} }
PERF_TEST_P_(DNNTestNetwork, OpenPose_pose_mpi_faster_4_stages) PERF_TEST_P_(DNNTestNetwork, OpenPose_pose_mpi_faster_4_stages)
{ {
if (backend == DNN_BACKEND_HALIDE) throw SkipTestException(""); if (backend == DNN_BACKEND_HALIDE ||
backend == DNN_BACKEND_INFERENCE_ENGINE && target == DNN_TARGET_MYRIAD)
throw SkipTestException("");
// The same .caffemodel but modified .prototxt // The same .caffemodel but modified .prototxt
// See https://github.com/CMU-Perceptual-Computing-Lab/openpose/blob/master/src/openpose/pose/poseParameters.cpp // See https://github.com/CMU-Perceptual-Computing-Lab/openpose/blob/master/src/openpose/pose/poseParameters.cpp
processNet("dnn/openpose_pose_mpi.caffemodel", "dnn/openpose_pose_mpi_faster_4_stages.prototxt", "", processNet("dnn/openpose_pose_mpi.caffemodel", "dnn/openpose_pose_mpi_faster_4_stages.prototxt", "",
...@@ -197,8 +229,7 @@ PERF_TEST_P_(DNNTestNetwork, OpenPose_pose_mpi_faster_4_stages) ...@@ -197,8 +229,7 @@ PERF_TEST_P_(DNNTestNetwork, OpenPose_pose_mpi_faster_4_stages)
PERF_TEST_P_(DNNTestNetwork, opencv_face_detector) PERF_TEST_P_(DNNTestNetwork, opencv_face_detector)
{ {
if (backend == DNN_BACKEND_HALIDE || if (backend == DNN_BACKEND_HALIDE)
backend == DNN_BACKEND_INFERENCE_ENGINE && target != DNN_TARGET_CPU)
throw SkipTestException(""); throw SkipTestException("");
processNet("dnn/opencv_face_detector.caffemodel", "dnn/opencv_face_detector.prototxt", "", processNet("dnn/opencv_face_detector.caffemodel", "dnn/opencv_face_detector.prototxt", "",
Mat(cv::Size(300, 300), CV_32FC3)); Mat(cv::Size(300, 300), CV_32FC3));
...@@ -207,7 +238,8 @@ PERF_TEST_P_(DNNTestNetwork, opencv_face_detector) ...@@ -207,7 +238,8 @@ PERF_TEST_P_(DNNTestNetwork, opencv_face_detector)
PERF_TEST_P_(DNNTestNetwork, Inception_v2_SSD_TensorFlow) PERF_TEST_P_(DNNTestNetwork, Inception_v2_SSD_TensorFlow)
{ {
if (backend == DNN_BACKEND_HALIDE || if (backend == DNN_BACKEND_HALIDE ||
backend == DNN_BACKEND_INFERENCE_ENGINE && target != DNN_TARGET_CPU) (backend == DNN_BACKEND_INFERENCE_ENGINE && target == DNN_TARGET_OPENCL) ||
(backend == DNN_BACKEND_INFERENCE_ENGINE && target == DNN_TARGET_OPENCL_FP16))
throw SkipTestException(""); throw SkipTestException("");
processNet("dnn/ssd_inception_v2_coco_2017_11_17.pb", "ssd_inception_v2_coco_2017_11_17.pbtxt", "", processNet("dnn/ssd_inception_v2_coco_2017_11_17.pb", "ssd_inception_v2_coco_2017_11_17.pbtxt", "",
Mat(cv::Size(300, 300), CV_32FC3)); Mat(cv::Size(300, 300), CV_32FC3));
...@@ -215,7 +247,8 @@ PERF_TEST_P_(DNNTestNetwork, Inception_v2_SSD_TensorFlow) ...@@ -215,7 +247,8 @@ PERF_TEST_P_(DNNTestNetwork, Inception_v2_SSD_TensorFlow)
PERF_TEST_P_(DNNTestNetwork, YOLOv3) PERF_TEST_P_(DNNTestNetwork, YOLOv3)
{ {
if (backend != DNN_BACKEND_DEFAULT) if (backend == DNN_BACKEND_HALIDE ||
backend == DNN_BACKEND_INFERENCE_ENGINE && target == DNN_TARGET_MYRIAD)
throw SkipTestException(""); throw SkipTestException("");
Mat sample = imread(findDataFile("dnn/dog416.png", false)); Mat sample = imread(findDataFile("dnn/dog416.png", false));
Mat inp; Mat inp;
...@@ -232,6 +265,7 @@ const tuple<DNNBackend, DNNTarget> testCases[] = { ...@@ -232,6 +265,7 @@ const tuple<DNNBackend, DNNTarget> testCases[] = {
tuple<DNNBackend, DNNTarget>(DNN_BACKEND_INFERENCE_ENGINE, DNN_TARGET_CPU), tuple<DNNBackend, DNNTarget>(DNN_BACKEND_INFERENCE_ENGINE, DNN_TARGET_CPU),
tuple<DNNBackend, DNNTarget>(DNN_BACKEND_INFERENCE_ENGINE, DNN_TARGET_OPENCL), tuple<DNNBackend, DNNTarget>(DNN_BACKEND_INFERENCE_ENGINE, DNN_TARGET_OPENCL),
tuple<DNNBackend, DNNTarget>(DNN_BACKEND_INFERENCE_ENGINE, DNN_TARGET_OPENCL_FP16), tuple<DNNBackend, DNNTarget>(DNN_BACKEND_INFERENCE_ENGINE, DNN_TARGET_OPENCL_FP16),
tuple<DNNBackend, DNNTarget>(DNN_BACKEND_INFERENCE_ENGINE, DNN_TARGET_MYRIAD),
#endif #endif
tuple<DNNBackend, DNNTarget>(DNN_BACKEND_DEFAULT, DNN_TARGET_CPU), tuple<DNNBackend, DNNTarget>(DNN_BACKEND_DEFAULT, DNN_TARGET_CPU),
tuple<DNNBackend, DNNTarget>(DNN_BACKEND_DEFAULT, DNN_TARGET_OPENCL), tuple<DNNBackend, DNNTarget>(DNN_BACKEND_DEFAULT, DNN_TARGET_OPENCL),
......
...@@ -288,7 +288,7 @@ namespace cv { ...@@ -288,7 +288,7 @@ namespace cv {
permute_params.set("order", paramOrder); permute_params.set("order", paramOrder);
darknet::LayerParameter lp; darknet::LayerParameter lp;
std::string layer_name = cv::format("premute_%d", layer_id); std::string layer_name = cv::format("permute_%d", layer_id);
lp.layer_name = layer_name; lp.layer_name = layer_name;
lp.layer_type = permute_params.type; lp.layer_type = permute_params.type;
lp.layerParams = permute_params; lp.layerParams = permute_params;
......
...@@ -1182,7 +1182,9 @@ struct Net::Impl ...@@ -1182,7 +1182,9 @@ struct Net::Impl
for (it = layers.begin(); it != layers.end(); ++it) for (it = layers.begin(); it != layers.end(); ++it)
{ {
LayerData &ld = it->second; LayerData &ld = it->second;
bool fused = ld.skip && ld.id != 0; if (ld.id == 0)
continue;
bool fused = ld.skip;
Ptr<Layer> layer = ld.layerInstance; Ptr<Layer> layer = ld.layerInstance;
if (!layer->supportBackend(preferableBackend)) if (!layer->supportBackend(preferableBackend))
...@@ -1259,7 +1261,7 @@ struct Net::Impl ...@@ -1259,7 +1261,7 @@ struct Net::Impl
CV_Assert(!ieNode.empty()); CV_Assert(!ieNode.empty());
ieNode->net = net; ieNode->net = net;
if (preferableTarget == DNN_TARGET_OPENCL_FP16 && !fused) if ((preferableTarget == DNN_TARGET_OPENCL_FP16 || preferableTarget == DNN_TARGET_MYRIAD) && !fused)
{ {
ieNode->layer->precision = InferenceEngine::Precision::FP16; ieNode->layer->precision = InferenceEngine::Precision::FP16;
auto weightableLayer = std::dynamic_pointer_cast<InferenceEngine::WeightableLayer>(ieNode->layer); auto weightableLayer = std::dynamic_pointer_cast<InferenceEngine::WeightableLayer>(ieNode->layer);
......
...@@ -117,7 +117,7 @@ public: ...@@ -117,7 +117,7 @@ public:
{ {
return backendId == DNN_BACKEND_DEFAULT || return backendId == DNN_BACKEND_DEFAULT ||
backendId == DNN_BACKEND_HALIDE && haveHalide() || backendId == DNN_BACKEND_HALIDE && haveHalide() ||
backendId == DNN_BACKEND_INFERENCE_ENGINE && haveInfEngine() && this->type != "Sigmoid"; backendId == DNN_BACKEND_INFERENCE_ENGINE && haveInfEngine();
} }
virtual Ptr<BackendNode> tryAttach(const Ptr<BackendNode>& node) CV_OVERRIDE virtual Ptr<BackendNode> tryAttach(const Ptr<BackendNode>& node) CV_OVERRIDE
...@@ -334,6 +334,7 @@ struct ReLUFunctor ...@@ -334,6 +334,7 @@ struct ReLUFunctor
lp.type = "ReLU"; lp.type = "ReLU";
std::shared_ptr<InferenceEngine::ReLULayer> ieLayer(new InferenceEngine::ReLULayer(lp)); std::shared_ptr<InferenceEngine::ReLULayer> ieLayer(new InferenceEngine::ReLULayer(lp));
ieLayer->negative_slope = slope; ieLayer->negative_slope = slope;
ieLayer->params["negative_slope"] = format("%f", slope);
return ieLayer; return ieLayer;
} }
#endif // HAVE_INF_ENGINE #endif // HAVE_INF_ENGINE
...@@ -431,6 +432,8 @@ struct ReLU6Functor ...@@ -431,6 +432,8 @@ struct ReLU6Functor
std::shared_ptr<InferenceEngine::ClampLayer> ieLayer(new InferenceEngine::ClampLayer(lp)); std::shared_ptr<InferenceEngine::ClampLayer> ieLayer(new InferenceEngine::ClampLayer(lp));
ieLayer->min_value = minValue; ieLayer->min_value = minValue;
ieLayer->max_value = maxValue; ieLayer->max_value = maxValue;
ieLayer->params["min"] = format("%f", minValue);
ieLayer->params["max"] = format("%f", maxValue);
return ieLayer; return ieLayer;
} }
#endif // HAVE_INF_ENGINE #endif // HAVE_INF_ENGINE
...@@ -556,8 +559,9 @@ struct SigmoidFunctor ...@@ -556,8 +559,9 @@ struct SigmoidFunctor
#ifdef HAVE_INF_ENGINE #ifdef HAVE_INF_ENGINE
InferenceEngine::CNNLayerPtr initInfEngine(InferenceEngine::LayerParams& lp) InferenceEngine::CNNLayerPtr initInfEngine(InferenceEngine::LayerParams& lp)
{ {
CV_Error(Error::StsNotImplemented, "Sigmoid"); lp.type = "Sigmoid";
return InferenceEngine::CNNLayerPtr(); std::shared_ptr<InferenceEngine::CNNLayer> ieLayer(new InferenceEngine::CNNLayer(lp));
return ieLayer;
} }
#endif // HAVE_INF_ENGINE #endif // HAVE_INF_ENGINE
......
...@@ -271,7 +271,7 @@ public: ...@@ -271,7 +271,7 @@ public:
virtual bool supportBackend(int backendId) CV_OVERRIDE virtual bool supportBackend(int backendId) CV_OVERRIDE
{ {
return backendId == DNN_BACKEND_DEFAULT || return backendId == DNN_BACKEND_DEFAULT ||
backendId == DNN_BACKEND_INFERENCE_ENGINE && haveInfEngine() && !_explicitSizes; backendId == DNN_BACKEND_INFERENCE_ENGINE && haveInfEngine();
} }
bool getMemoryShapes(const std::vector<MatShape> &inputs, bool getMemoryShapes(const std::vector<MatShape> &inputs,
...@@ -484,18 +484,33 @@ public: ...@@ -484,18 +484,33 @@ public:
#ifdef HAVE_INF_ENGINE #ifdef HAVE_INF_ENGINE
InferenceEngine::LayerParams lp; InferenceEngine::LayerParams lp;
lp.name = name; lp.name = name;
lp.type = "PriorBox"; lp.type = _explicitSizes ? "PriorBoxClustered" : "PriorBox";
lp.precision = InferenceEngine::Precision::FP32; lp.precision = InferenceEngine::Precision::FP32;
std::shared_ptr<InferenceEngine::CNNLayer> ieLayer(new InferenceEngine::CNNLayer(lp)); std::shared_ptr<InferenceEngine::CNNLayer> ieLayer(new InferenceEngine::CNNLayer(lp));
ieLayer->params["min_size"] = format("%f", _minSize); if (_explicitSizes)
ieLayer->params["max_size"] = _maxSize > 0 ? format("%f", _maxSize) : ""; {
CV_Assert(!_boxWidths.empty(), !_boxHeights.empty(),
if (!_aspectRatios.empty()) _boxWidths.size() == _boxHeights.size());
ieLayer->params["width"] = format("%f", _boxWidths[0]);
ieLayer->params["height"] = format("%f", _boxHeights[0]);
for (int i = 1; i < _boxWidths.size(); ++i)
{
ieLayer->params["width"] += format(",%f", _boxWidths[i]);
ieLayer->params["height"] += format(",%f", _boxHeights[i]);
}
}
else
{ {
ieLayer->params["aspect_ratio"] = format("%f", _aspectRatios[0]); ieLayer->params["min_size"] = format("%f", _minSize);
for (int i = 1; i < _aspectRatios.size(); ++i) ieLayer->params["max_size"] = _maxSize > 0 ? format("%f", _maxSize) : "";
ieLayer->params["aspect_ratio"] += format(",%f", _aspectRatios[i]);
if (!_aspectRatios.empty())
{
ieLayer->params["aspect_ratio"] = format("%f", _aspectRatios[0]);
for (int i = 1; i < _aspectRatios.size(); ++i)
ieLayer->params["aspect_ratio"] += format(",%f", _aspectRatios[i]);
}
} }
ieLayer->params["flip"] = "0"; // We already flipped aspect ratios. ieLayer->params["flip"] = "0"; // We already flipped aspect ratios.
......
...@@ -95,11 +95,6 @@ public: ...@@ -95,11 +95,6 @@ public:
return false; return false;
} }
virtual bool supportBackend(int backendId) CV_OVERRIDE
{
return backendId == DNN_BACKEND_DEFAULT;
}
float logistic_activate(float x) { return 1.F / (1.F + exp(-x)); } float logistic_activate(float x) { return 1.F / (1.F + exp(-x)); }
void softmax_activate(const float* input, const int n, const float temp, float* output) void softmax_activate(const float* input, const int n, const float temp, float* output)
......
...@@ -6,6 +6,7 @@ ...@@ -6,6 +6,7 @@
// Third party copyrights are property of their respective owners. // Third party copyrights are property of their respective owners.
#include "../precomp.hpp" #include "../precomp.hpp"
#include "layers_common.hpp" #include "layers_common.hpp"
#include "../op_inf_engine.hpp"
#include <opencv2/imgproc.hpp> #include <opencv2/imgproc.hpp>
namespace cv { namespace dnn { namespace cv { namespace dnn {
...@@ -39,6 +40,12 @@ public: ...@@ -39,6 +40,12 @@ public:
return (outputs[0][2] == inputs[0][2]) && (outputs[0][3] == inputs[0][3]); return (outputs[0][2] == inputs[0][2]) && (outputs[0][3] == inputs[0][3]);
} }
virtual bool supportBackend(int backendId) CV_OVERRIDE
{
return backendId == DNN_BACKEND_DEFAULT ||
backendId == DNN_BACKEND_INFERENCE_ENGINE && haveInfEngine();
}
virtual void finalize(const std::vector<Mat*>& inputs, std::vector<Mat> &outputs) CV_OVERRIDE virtual void finalize(const std::vector<Mat*>& inputs, std::vector<Mat> &outputs) CV_OVERRIDE
{ {
if (!outWidth && !outHeight) if (!outWidth && !outHeight)
...@@ -75,6 +82,26 @@ public: ...@@ -75,6 +82,26 @@ public:
} }
} }
} }
virtual Ptr<BackendNode> initInfEngine(const std::vector<Ptr<BackendWrapper> >&) CV_OVERRIDE
{
#ifdef HAVE_INF_ENGINE
InferenceEngine::LayerParams lp;
lp.name = name;
lp.type = "Resample";
lp.precision = InferenceEngine::Precision::FP32;
std::shared_ptr<InferenceEngine::CNNLayer> ieLayer(new InferenceEngine::CNNLayer(lp));
ieLayer->params["type"] = "caffe.ResampleParameter.NEAREST";
ieLayer->params["antialias"] = "0";
ieLayer->params["width"] = cv::format("%d", outWidth);
ieLayer->params["height"] = cv::format("%d", outHeight);
return Ptr<BackendNode>(new InfEngineBackendNode(ieLayer));
#endif // HAVE_INF_ENGINE
return Ptr<BackendNode>();
}
private: private:
int outWidth, outHeight, zoomFactor; int outWidth, outHeight, zoomFactor;
bool alignCorners; bool alignCorners;
......
...@@ -18,11 +18,6 @@ namespace cv { namespace dnn { ...@@ -18,11 +18,6 @@ namespace cv { namespace dnn {
#ifdef HAVE_INF_ENGINE #ifdef HAVE_INF_ENGINE
static int infEngineVersion()
{
return std::atoi(InferenceEngine::GetInferenceEngineVersion()->buildNumber);
}
InfEngineBackendNode::InfEngineBackendNode(const InferenceEngine::CNNLayerPtr& _layer) InfEngineBackendNode::InfEngineBackendNode(const InferenceEngine::CNNLayerPtr& _layer)
: BackendNode(DNN_BACKEND_INFERENCE_ENGINE), layer(_layer) {} : BackendNode(DNN_BACKEND_INFERENCE_ENGINE), layer(_layer) {}
...@@ -59,27 +54,23 @@ infEngineWrappers(const std::vector<Ptr<BackendWrapper> >& ptrs) ...@@ -59,27 +54,23 @@ infEngineWrappers(const std::vector<Ptr<BackendWrapper> >& ptrs)
return wrappers; return wrappers;
} }
static InferenceEngine::Layout estimateLayout(const Mat& m)
{
if (m.dims == 4)
return InferenceEngine::Layout::NCHW;
else if (m.dims == 2)
return InferenceEngine::Layout::NC;
else
return InferenceEngine::Layout::ANY;
}
static InferenceEngine::DataPtr wrapToInfEngineDataNode(const Mat& m, const std::string& name = "") static InferenceEngine::DataPtr wrapToInfEngineDataNode(const Mat& m, const std::string& name = "")
{ {
std::vector<size_t> reversedShape(&m.size[0], &m.size[0] + m.dims); std::vector<size_t> reversedShape(&m.size[0], &m.size[0] + m.dims);
std::reverse(reversedShape.begin(), reversedShape.end()); std::reverse(reversedShape.begin(), reversedShape.end());
if (infEngineVersion() > 5855) return InferenceEngine::DataPtr(
{ new InferenceEngine::Data(name, reversedShape, InferenceEngine::Precision::FP32, estimateLayout(m))
InferenceEngine::Layout l = InferenceEngine::Layout::ANY; );
if (m.dims == 4)
l = InferenceEngine::Layout::NCHW;
else if (m.dims == 2)
l = InferenceEngine::Layout::NC;
return InferenceEngine::DataPtr(
new InferenceEngine::Data(name, reversedShape, InferenceEngine::Precision::FP32, l)
);
}
else
{
return InferenceEngine::DataPtr(
new InferenceEngine::Data(name, reversedShape, InferenceEngine::Precision::FP32)
);
}
} }
InferenceEngine::TBlob<float>::Ptr wrapToInfEngineBlob(const Mat& m, const std::vector<size_t>& shape, InferenceEngine::TBlob<float>::Ptr wrapToInfEngineBlob(const Mat& m, const std::vector<size_t>& shape,
...@@ -108,7 +99,7 @@ InfEngineBackendWrapper::InfEngineBackendWrapper(int targetId, const cv::Mat& m) ...@@ -108,7 +99,7 @@ InfEngineBackendWrapper::InfEngineBackendWrapper(int targetId, const cv::Mat& m)
: BackendWrapper(DNN_BACKEND_INFERENCE_ENGINE, targetId) : BackendWrapper(DNN_BACKEND_INFERENCE_ENGINE, targetId)
{ {
dataPtr = wrapToInfEngineDataNode(m); dataPtr = wrapToInfEngineDataNode(m);
blob = wrapToInfEngineBlob(m); blob = wrapToInfEngineBlob(m, estimateLayout(m));
} }
InfEngineBackendWrapper::~InfEngineBackendWrapper() InfEngineBackendWrapper::~InfEngineBackendWrapper()
...@@ -252,7 +243,8 @@ InfEngineBackendNet::getLayerByName(const char *layerName, InferenceEngine::CNNL ...@@ -252,7 +243,8 @@ InfEngineBackendNet::getLayerByName(const char *layerName, InferenceEngine::CNNL
void InfEngineBackendNet::setTargetDevice(InferenceEngine::TargetDevice device) noexcept void InfEngineBackendNet::setTargetDevice(InferenceEngine::TargetDevice device) noexcept
{ {
if (device != InferenceEngine::TargetDevice::eCPU && if (device != InferenceEngine::TargetDevice::eCPU &&
device != InferenceEngine::TargetDevice::eGPU) device != InferenceEngine::TargetDevice::eGPU &&
device != InferenceEngine::TargetDevice::eMYRIAD)
CV_Error(Error::StsNotImplemented, ""); CV_Error(Error::StsNotImplemented, "");
targetDevice = device; targetDevice = device;
} }
...@@ -352,6 +344,11 @@ void InfEngineBackendNet::init(int targetId) ...@@ -352,6 +344,11 @@ void InfEngineBackendNet::init(int targetId)
case DNN_TARGET_CPU: setTargetDevice(InferenceEngine::TargetDevice::eCPU); break; case DNN_TARGET_CPU: setTargetDevice(InferenceEngine::TargetDevice::eCPU); break;
case DNN_TARGET_OPENCL_FP16: setPrecision(InferenceEngine::Precision::FP16); // Fallback to the next. case DNN_TARGET_OPENCL_FP16: setPrecision(InferenceEngine::Precision::FP16); // Fallback to the next.
case DNN_TARGET_OPENCL: setTargetDevice(InferenceEngine::TargetDevice::eGPU); break; case DNN_TARGET_OPENCL: setTargetDevice(InferenceEngine::TargetDevice::eGPU); break;
case DNN_TARGET_MYRIAD:
{
setPrecision(InferenceEngine::Precision::FP16);
setTargetDevice(InferenceEngine::TargetDevice::eMYRIAD); break;
}
default: default:
CV_Error(Error::StsError, format("Unknown target identifier: %d", targetId)); CV_Error(Error::StsError, format("Unknown target identifier: %d", targetId));
} }
...@@ -368,7 +365,7 @@ void InfEngineBackendNet::initPlugin(InferenceEngine::ICNNNetwork& net) ...@@ -368,7 +365,7 @@ void InfEngineBackendNet::initPlugin(InferenceEngine::ICNNNetwork& net)
InferenceEngine::ResponseDesc resp; InferenceEngine::ResponseDesc resp;
plugin = InferenceEngine::PluginDispatcher({""}).getSuitablePlugin(targetDevice); plugin = InferenceEngine::PluginDispatcher({""}).getSuitablePlugin(targetDevice);
if (infEngineVersion() > 5855 && targetDevice == InferenceEngine::TargetDevice::eCPU) if (targetDevice == InferenceEngine::TargetDevice::eCPU)
{ {
#ifdef _WIN32 #ifdef _WIN32
InferenceEngine::IExtensionPtr extension = InferenceEngine::IExtensionPtr extension =
......
...@@ -49,7 +49,14 @@ public: ...@@ -49,7 +49,14 @@ public:
throw SkipTestException("OpenCL is not available/disabled in OpenCV"); throw SkipTestException("OpenCL is not available/disabled in OpenCV");
} }
} }
if (target == DNN_TARGET_OPENCL_FP16) if (backend == DNN_BACKEND_INFERENCE_ENGINE && target == DNN_TARGET_MYRIAD)
{
if (!checkMyriadTarget())
{
throw SkipTestException("Myriad is not available/disabled in OpenCV");
}
}
if (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD)
{ {
l1 = l1 == 0.0 ? 4e-3 : l1; l1 = l1 == 0.0 ? 4e-3 : l1;
lInf = lInf == 0.0 ? 2e-2 : lInf; lInf = lInf == 0.0 ? 2e-2 : lInf;
...@@ -80,10 +87,7 @@ public: ...@@ -80,10 +87,7 @@ public:
} }
Mat out = net.forward(outputLayer).clone(); Mat out = net.forward(outputLayer).clone();
if (outputLayer == "detection_out") check(outDefault, out, outputLayer, l1, lInf, "First run");
normAssertDetections(outDefault, out, "First run", 0.2, l1, lInf);
else
normAssert(outDefault, out, "First run", l1, lInf);
// Test 2: change input. // Test 2: change input.
float* inpData = (float*)inp.data; float* inpData = (float*)inp.data;
...@@ -97,18 +101,33 @@ public: ...@@ -97,18 +101,33 @@ public:
net.setInput(inp); net.setInput(inp);
outDefault = netDefault.forward(outputLayer).clone(); outDefault = netDefault.forward(outputLayer).clone();
out = net.forward(outputLayer).clone(); out = net.forward(outputLayer).clone();
check(outDefault, out, outputLayer, l1, lInf, "Second run");
}
void check(Mat& ref, Mat& out, const std::string& outputLayer, double l1, double lInf, const char* msg)
{
if (outputLayer == "detection_out") if (outputLayer == "detection_out")
normAssertDetections(outDefault, out, "Second run", 0.2, l1, lInf); {
if (backend == DNN_BACKEND_INFERENCE_ENGINE)
{
// Inference Engine produces detections terminated by a row which starts from -1.
out = out.reshape(1, out.total() / 7);
int numDetections = 0;
while (numDetections < out.rows && out.at<float>(numDetections, 0) != -1)
{
numDetections += 1;
}
out = out.rowRange(0, numDetections);
}
normAssertDetections(ref, out, msg, 0.2, l1, lInf);
}
else else
normAssert(outDefault, out, "Second run", l1, lInf); normAssert(ref, out, msg, l1, lInf);
} }
}; };
TEST_P(DNNTestNetwork, AlexNet) TEST_P(DNNTestNetwork, AlexNet)
{ {
if (backend == DNN_BACKEND_INFERENCE_ENGINE && target != DNN_TARGET_CPU)
throw SkipTestException("");
processNet("dnn/bvlc_alexnet.caffemodel", "dnn/bvlc_alexnet.prototxt", processNet("dnn/bvlc_alexnet.caffemodel", "dnn/bvlc_alexnet.prototxt",
Size(227, 227), "prob", Size(227, 227), "prob",
target == DNN_TARGET_OPENCL ? "dnn/halide_scheduler_opencl_alexnet.yml" : target == DNN_TARGET_OPENCL ? "dnn/halide_scheduler_opencl_alexnet.yml" :
...@@ -158,8 +177,7 @@ TEST_P(DNNTestNetwork, ENet) ...@@ -158,8 +177,7 @@ TEST_P(DNNTestNetwork, ENet)
TEST_P(DNNTestNetwork, MobileNet_SSD_Caffe) TEST_P(DNNTestNetwork, MobileNet_SSD_Caffe)
{ {
if (backend == DNN_BACKEND_HALIDE || if (backend == DNN_BACKEND_HALIDE)
backend == DNN_BACKEND_INFERENCE_ENGINE && target != DNN_TARGET_CPU)
throw SkipTestException(""); throw SkipTestException("");
Mat sample = imread(findDataFile("dnn/street.png", false)); Mat sample = imread(findDataFile("dnn/street.png", false));
Mat inp = blobFromImage(sample, 1.0f / 127.5, Size(300, 300), Scalar(127.5, 127.5, 127.5), false); Mat inp = blobFromImage(sample, 1.0f / 127.5, Size(300, 300), Scalar(127.5, 127.5, 127.5), false);
...@@ -170,10 +188,11 @@ TEST_P(DNNTestNetwork, MobileNet_SSD_Caffe) ...@@ -170,10 +188,11 @@ TEST_P(DNNTestNetwork, MobileNet_SSD_Caffe)
inp, "detection_out", "", l1, lInf); inp, "detection_out", "", l1, lInf);
} }
// TODO: update MobileNet model.
TEST_P(DNNTestNetwork, MobileNet_SSD_TensorFlow) TEST_P(DNNTestNetwork, MobileNet_SSD_TensorFlow)
{ {
if (backend == DNN_BACKEND_HALIDE || if (backend == DNN_BACKEND_HALIDE ||
backend == DNN_BACKEND_INFERENCE_ENGINE && target != DNN_TARGET_CPU) backend == DNN_BACKEND_INFERENCE_ENGINE)
throw SkipTestException(""); throw SkipTestException("");
Mat sample = imread(findDataFile("dnn/street.png", false)); Mat sample = imread(findDataFile("dnn/street.png", false));
Mat inp = blobFromImage(sample, 1.0f / 127.5, Size(300, 300), Scalar(127.5, 127.5, 127.5), false); Mat inp = blobFromImage(sample, 1.0f / 127.5, Size(300, 300), Scalar(127.5, 127.5, 127.5), false);
...@@ -185,31 +204,38 @@ TEST_P(DNNTestNetwork, MobileNet_SSD_TensorFlow) ...@@ -185,31 +204,38 @@ TEST_P(DNNTestNetwork, MobileNet_SSD_TensorFlow)
TEST_P(DNNTestNetwork, SSD_VGG16) TEST_P(DNNTestNetwork, SSD_VGG16)
{ {
if ((backend == DNN_BACKEND_DEFAULT && target == DNN_TARGET_OPENCL_FP16) || if (backend == DNN_BACKEND_HALIDE && target == DNN_TARGET_CPU)
(backend == DNN_BACKEND_HALIDE && target == DNN_TARGET_CPU) ||
(backend == DNN_BACKEND_INFERENCE_ENGINE && target != DNN_TARGET_CPU))
throw SkipTestException(""); throw SkipTestException("");
double scoreThreshold = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.0252 : 0.0;
Mat sample = imread(findDataFile("dnn/street.png", false));
Mat inp = blobFromImage(sample, 1.0f, Size(300, 300), Scalar(), false);
processNet("dnn/VGG_ILSVRC2016_SSD_300x300_iter_440000.caffemodel", processNet("dnn/VGG_ILSVRC2016_SSD_300x300_iter_440000.caffemodel",
"dnn/ssd_vgg16.prototxt", Size(300, 300), "detection_out"); "dnn/ssd_vgg16.prototxt", inp, "detection_out", "", scoreThreshold);
} }
TEST_P(DNNTestNetwork, OpenPose_pose_coco) TEST_P(DNNTestNetwork, OpenPose_pose_coco)
{ {
if (backend == DNN_BACKEND_HALIDE) throw SkipTestException(""); if (backend == DNN_BACKEND_HALIDE ||
backend == DNN_BACKEND_INFERENCE_ENGINE && target == DNN_TARGET_MYRIAD)
throw SkipTestException("");
processNet("dnn/openpose_pose_coco.caffemodel", "dnn/openpose_pose_coco.prototxt", processNet("dnn/openpose_pose_coco.caffemodel", "dnn/openpose_pose_coco.prototxt",
Size(368, 368)); Size(368, 368));
} }
TEST_P(DNNTestNetwork, OpenPose_pose_mpi) TEST_P(DNNTestNetwork, OpenPose_pose_mpi)
{ {
if (backend == DNN_BACKEND_HALIDE) throw SkipTestException(""); if (backend == DNN_BACKEND_HALIDE ||
backend == DNN_BACKEND_INFERENCE_ENGINE && target == DNN_TARGET_MYRIAD)
throw SkipTestException("");
processNet("dnn/openpose_pose_mpi.caffemodel", "dnn/openpose_pose_mpi.prototxt", processNet("dnn/openpose_pose_mpi.caffemodel", "dnn/openpose_pose_mpi.prototxt",
Size(368, 368)); Size(368, 368));
} }
TEST_P(DNNTestNetwork, OpenPose_pose_mpi_faster_4_stages) TEST_P(DNNTestNetwork, OpenPose_pose_mpi_faster_4_stages)
{ {
if (backend == DNN_BACKEND_HALIDE) throw SkipTestException(""); if (backend == DNN_BACKEND_HALIDE ||
backend == DNN_BACKEND_INFERENCE_ENGINE && target == DNN_TARGET_MYRIAD)
throw SkipTestException("");
// The same .caffemodel but modified .prototxt // The same .caffemodel but modified .prototxt
// See https://github.com/CMU-Perceptual-Computing-Lab/openpose/blob/master/src/openpose/pose/poseParameters.cpp // See https://github.com/CMU-Perceptual-Computing-Lab/openpose/blob/master/src/openpose/pose/poseParameters.cpp
processNet("dnn/openpose_pose_mpi.caffemodel", "dnn/openpose_pose_mpi_faster_4_stages.prototxt", processNet("dnn/openpose_pose_mpi.caffemodel", "dnn/openpose_pose_mpi_faster_4_stages.prototxt",
...@@ -226,11 +252,13 @@ TEST_P(DNNTestNetwork, OpenFace) ...@@ -226,11 +252,13 @@ TEST_P(DNNTestNetwork, OpenFace)
TEST_P(DNNTestNetwork, opencv_face_detector) TEST_P(DNNTestNetwork, opencv_face_detector)
{ {
if (backend == DNN_BACKEND_HALIDE || if (backend == DNN_BACKEND_HALIDE)
backend == DNN_BACKEND_INFERENCE_ENGINE && target != DNN_TARGET_CPU)
throw SkipTestException(""); throw SkipTestException("");
Size inpSize;
if (backend == DNN_BACKEND_INFERENCE_ENGINE && target == DNN_TARGET_MYRIAD)
inpSize = Size(300, 300);
Mat img = imread(findDataFile("gpu/lbpcascade/er.png", false)); Mat img = imread(findDataFile("gpu/lbpcascade/er.png", false));
Mat inp = blobFromImage(img, 1.0, Size(), Scalar(104.0, 177.0, 123.0), false, false); Mat inp = blobFromImage(img, 1.0, inpSize, Scalar(104.0, 177.0, 123.0), false, false);
processNet("dnn/opencv_face_detector.caffemodel", "dnn/opencv_face_detector.prototxt", processNet("dnn/opencv_face_detector.caffemodel", "dnn/opencv_face_detector.prototxt",
inp, "detection_out"); inp, "detection_out");
} }
...@@ -238,12 +266,13 @@ TEST_P(DNNTestNetwork, opencv_face_detector) ...@@ -238,12 +266,13 @@ TEST_P(DNNTestNetwork, opencv_face_detector)
TEST_P(DNNTestNetwork, Inception_v2_SSD_TensorFlow) TEST_P(DNNTestNetwork, Inception_v2_SSD_TensorFlow)
{ {
if (backend == DNN_BACKEND_HALIDE || if (backend == DNN_BACKEND_HALIDE ||
backend == DNN_BACKEND_INFERENCE_ENGINE && target != DNN_TARGET_CPU) (backend == DNN_BACKEND_INFERENCE_ENGINE && target == DNN_TARGET_OPENCL) ||
(backend == DNN_BACKEND_INFERENCE_ENGINE && target == DNN_TARGET_OPENCL_FP16))
throw SkipTestException(""); throw SkipTestException("");
Mat sample = imread(findDataFile("dnn/street.png", false)); Mat sample = imread(findDataFile("dnn/street.png", false));
Mat inp = blobFromImage(sample, 1.0f / 127.5, Size(300, 300), Scalar(127.5, 127.5, 127.5), false); Mat inp = blobFromImage(sample, 1.0f / 127.5, Size(300, 300), Scalar(127.5, 127.5, 127.5), false);
float l1 = (backend == DNN_BACKEND_DEFAULT && target == DNN_TARGET_OPENCL_FP16) ? 0.008 : 0.0; float l1 = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.008 : 0.0;
float lInf = (backend == DNN_BACKEND_DEFAULT && target == DNN_TARGET_OPENCL_FP16) ? 0.07 : 0.0; float lInf = (target == DNN_TARGET_OPENCL_FP16 || target == DNN_TARGET_MYRIAD) ? 0.07 : 0.0;
processNet("dnn/ssd_inception_v2_coco_2017_11_17.pb", "dnn/ssd_inception_v2_coco_2017_11_17.pbtxt", processNet("dnn/ssd_inception_v2_coco_2017_11_17.pb", "dnn/ssd_inception_v2_coco_2017_11_17.pbtxt",
inp, "detection_out", "", l1, lInf); inp, "detection_out", "", l1, lInf);
} }
...@@ -252,7 +281,8 @@ TEST_P(DNNTestNetwork, DenseNet_121) ...@@ -252,7 +281,8 @@ TEST_P(DNNTestNetwork, DenseNet_121)
{ {
if ((backend == DNN_BACKEND_HALIDE) || if ((backend == DNN_BACKEND_HALIDE) ||
(backend == DNN_BACKEND_DEFAULT && target == DNN_TARGET_OPENCL_FP16) || (backend == DNN_BACKEND_DEFAULT && target == DNN_TARGET_OPENCL_FP16) ||
(backend == DNN_BACKEND_INFERENCE_ENGINE && target == DNN_TARGET_OPENCL_FP16)) (backend == DNN_BACKEND_INFERENCE_ENGINE && (target == DNN_TARGET_OPENCL_FP16 ||
target == DNN_TARGET_MYRIAD)))
throw SkipTestException(""); throw SkipTestException("");
processNet("dnn/DenseNet_121.caffemodel", "dnn/DenseNet_121.prototxt", Size(224, 224), "", "caffe"); processNet("dnn/DenseNet_121.caffemodel", "dnn/DenseNet_121.prototxt", Size(224, 224), "", "caffe");
} }
...@@ -266,6 +296,7 @@ const tuple<DNNBackend, DNNTarget> testCases[] = { ...@@ -266,6 +296,7 @@ const tuple<DNNBackend, DNNTarget> testCases[] = {
tuple<DNNBackend, DNNTarget>(DNN_BACKEND_INFERENCE_ENGINE, DNN_TARGET_CPU), tuple<DNNBackend, DNNTarget>(DNN_BACKEND_INFERENCE_ENGINE, DNN_TARGET_CPU),
tuple<DNNBackend, DNNTarget>(DNN_BACKEND_INFERENCE_ENGINE, DNN_TARGET_OPENCL), tuple<DNNBackend, DNNTarget>(DNN_BACKEND_INFERENCE_ENGINE, DNN_TARGET_OPENCL),
tuple<DNNBackend, DNNTarget>(DNN_BACKEND_INFERENCE_ENGINE, DNN_TARGET_OPENCL_FP16), tuple<DNNBackend, DNNTarget>(DNN_BACKEND_INFERENCE_ENGINE, DNN_TARGET_OPENCL_FP16),
tuple<DNNBackend, DNNTarget>(DNN_BACKEND_INFERENCE_ENGINE, DNN_TARGET_MYRIAD),
#endif #endif
tuple<DNNBackend, DNNTarget>(DNN_BACKEND_DEFAULT, DNN_TARGET_OPENCL), tuple<DNNBackend, DNNTarget>(DNN_BACKEND_DEFAULT, DNN_TARGET_OPENCL),
tuple<DNNBackend, DNNTarget>(DNN_BACKEND_DEFAULT, DNN_TARGET_OPENCL_FP16) tuple<DNNBackend, DNNTarget>(DNN_BACKEND_DEFAULT, DNN_TARGET_OPENCL_FP16)
......
...@@ -147,6 +147,28 @@ inline void normAssertDetections(cv::Mat ref, cv::Mat out, const char *comment = ...@@ -147,6 +147,28 @@ inline void normAssertDetections(cv::Mat ref, cv::Mat out, const char *comment =
testBoxes, comment, confThreshold, scores_diff, boxes_iou_diff); testBoxes, comment, confThreshold, scores_diff, boxes_iou_diff);
} }
inline bool checkMyriadTarget()
{
#ifndef HAVE_INF_ENGINE
return false;
#endif
cv::dnn::Net net;
cv::dnn::LayerParams lp;
net.addLayerToPrev("testLayer", "Identity", lp);
net.setPreferableBackend(cv::dnn::DNN_BACKEND_INFERENCE_ENGINE);
net.setPreferableTarget(cv::dnn::DNN_TARGET_MYRIAD);
net.setInput(cv::Mat::zeros(1, 1, CV_32FC1));
try
{
net.forward();
}
catch(...)
{
return false;
}
return true;
}
inline bool readFileInMemory(const std::string& filename, std::string& content) inline bool readFileInMemory(const std::string& filename, std::string& content)
{ {
std::ios::openmode mode = std::ios::in | std::ios::binary; std::ios::openmode mode = std::ios::in | std::ios::binary;
......
...@@ -71,13 +71,31 @@ static void testDarknetModel(const std::string& cfg, const std::string& weights, ...@@ -71,13 +71,31 @@ static void testDarknetModel(const std::string& cfg, const std::string& weights,
const std::vector<int>& refClassIds, const std::vector<int>& refClassIds,
const std::vector<float>& refConfidences, const std::vector<float>& refConfidences,
const std::vector<Rect2d>& refBoxes, const std::vector<Rect2d>& refBoxes,
int targetId, float confThreshold = 0.24) int backendId, int targetId, float scoreDiff = 0.0,
float iouDiff = 0.0, float confThreshold = 0.24)
{ {
if (backendId == DNN_BACKEND_DEFAULT && targetId == DNN_TARGET_OPENCL)
{
#ifdef HAVE_OPENCL
if (!cv::ocl::useOpenCL())
#endif
{
throw SkipTestException("OpenCL is not available/disabled in OpenCV");
}
}
if (backendId == DNN_BACKEND_INFERENCE_ENGINE && targetId == DNN_TARGET_MYRIAD)
{
if (!checkMyriadTarget())
{
throw SkipTestException("Myriad is not available/disabled in OpenCV");
}
}
Mat sample = imread(_tf("dog416.png")); Mat sample = imread(_tf("dog416.png"));
Mat inp = blobFromImage(sample, 1.0/255, Size(416, 416), Scalar(), true, false); Mat inp = blobFromImage(sample, 1.0/255, Size(416, 416), Scalar(), true, false);
Net net = readNet(findDataFile("dnn/" + cfg, false), Net net = readNet(findDataFile("dnn/" + cfg, false),
findDataFile("dnn/" + weights, false)); findDataFile("dnn/" + weights, false));
net.setPreferableBackend(backendId);
net.setPreferableTarget(targetId); net.setPreferableTarget(targetId);
net.setInput(inp); net.setInput(inp);
std::vector<Mat> outs; std::vector<Mat> outs;
...@@ -108,14 +126,17 @@ static void testDarknetModel(const std::string& cfg, const std::string& weights, ...@@ -108,14 +126,17 @@ static void testDarknetModel(const std::string& cfg, const std::string& weights,
} }
} }
normAssertDetections(refClassIds, refConfidences, refBoxes, classIds, normAssertDetections(refClassIds, refConfidences, refBoxes, classIds,
confidences, boxes, "", confThreshold, 8e-5, 3e-5); confidences, boxes, "", confThreshold, scoreDiff, iouDiff);
} }
typedef testing::TestWithParam<DNNTarget> Test_Darknet_nets; typedef testing::TestWithParam<tuple<DNNBackend, DNNTarget> > Test_Darknet_nets;
TEST_P(Test_Darknet_nets, YoloVoc) TEST_P(Test_Darknet_nets, YoloVoc)
{ {
int targetId = GetParam(); int backendId = get<0>(GetParam());
int targetId = get<1>(GetParam());
if (backendId == DNN_BACKEND_INFERENCE_ENGINE && targetId == DNN_TARGET_MYRIAD)
throw SkipTestException("");
std::vector<cv::String> outNames(1, "detection_out"); std::vector<cv::String> outNames(1, "detection_out");
std::vector<int> classIds(3); std::vector<int> classIds(3);
...@@ -124,26 +145,34 @@ TEST_P(Test_Darknet_nets, YoloVoc) ...@@ -124,26 +145,34 @@ TEST_P(Test_Darknet_nets, YoloVoc)
classIds[0] = 6; confidences[0] = 0.750469f; boxes[0] = Rect2d(0.577374, 0.127391, 0.325575, 0.173418); // a car classIds[0] = 6; confidences[0] = 0.750469f; boxes[0] = Rect2d(0.577374, 0.127391, 0.325575, 0.173418); // a car
classIds[1] = 1; confidences[1] = 0.780879f; boxes[1] = Rect2d(0.270762, 0.264102, 0.461713, 0.48131); // a bycicle classIds[1] = 1; confidences[1] = 0.780879f; boxes[1] = Rect2d(0.270762, 0.264102, 0.461713, 0.48131); // a bycicle
classIds[2] = 11; confidences[2] = 0.901615f; boxes[2] = Rect2d(0.1386, 0.338509, 0.282737, 0.60028); // a dog classIds[2] = 11; confidences[2] = 0.901615f; boxes[2] = Rect2d(0.1386, 0.338509, 0.282737, 0.60028); // a dog
double scoreDiff = (targetId == DNN_TARGET_OPENCL_FP16 || targetId == DNN_TARGET_MYRIAD) ? 7e-3 : 8e-5;
double iouDiff = (targetId == DNN_TARGET_OPENCL_FP16 || targetId == DNN_TARGET_MYRIAD) ? 0.013 : 3e-5;
testDarknetModel("yolo-voc.cfg", "yolo-voc.weights", outNames, testDarknetModel("yolo-voc.cfg", "yolo-voc.weights", outNames,
classIds, confidences, boxes, targetId); classIds, confidences, boxes, backendId, targetId, scoreDiff, iouDiff);
} }
TEST_P(Test_Darknet_nets, TinyYoloVoc) TEST_P(Test_Darknet_nets, TinyYoloVoc)
{ {
int targetId = GetParam(); int backendId = get<0>(GetParam());
int targetId = get<1>(GetParam());
std::vector<cv::String> outNames(1, "detection_out"); std::vector<cv::String> outNames(1, "detection_out");
std::vector<int> classIds(2); std::vector<int> classIds(2);
std::vector<float> confidences(2); std::vector<float> confidences(2);
std::vector<Rect2d> boxes(2); std::vector<Rect2d> boxes(2);
classIds[0] = 6; confidences[0] = 0.761967f; boxes[0] = Rect2d(0.579042, 0.159161, 0.31544, 0.160779); // a car classIds[0] = 6; confidences[0] = 0.761967f; boxes[0] = Rect2d(0.579042, 0.159161, 0.31544, 0.160779); // a car
classIds[1] = 11; confidences[1] = 0.780595f; boxes[1] = Rect2d(0.129696, 0.386467, 0.315579, 0.534527); // a dog classIds[1] = 11; confidences[1] = 0.780595f; boxes[1] = Rect2d(0.129696, 0.386467, 0.315579, 0.534527); // a dog
double scoreDiff = (targetId == DNN_TARGET_OPENCL_FP16 || targetId == DNN_TARGET_MYRIAD) ? 8e-3 : 8e-5;
double iouDiff = (targetId == DNN_TARGET_OPENCL_FP16 || targetId == DNN_TARGET_MYRIAD) ? 8e-3 : 3e-5;
testDarknetModel("tiny-yolo-voc.cfg", "tiny-yolo-voc.weights", outNames, testDarknetModel("tiny-yolo-voc.cfg", "tiny-yolo-voc.weights", outNames,
classIds, confidences, boxes, targetId); classIds, confidences, boxes, backendId, targetId, scoreDiff, iouDiff);
} }
TEST_P(Test_Darknet_nets, YOLOv3) TEST_P(Test_Darknet_nets, YOLOv3)
{ {
int targetId = GetParam(); int backendId = get<0>(GetParam());
int targetId = get<1>(GetParam());
if (backendId == DNN_BACKEND_INFERENCE_ENGINE && targetId == DNN_TARGET_MYRIAD)
throw SkipTestException("");
std::vector<cv::String> outNames(3); std::vector<cv::String> outNames(3);
outNames[0] = "yolo_82"; outNames[0] = "yolo_82";
outNames[1] = "yolo_94"; outNames[1] = "yolo_94";
...@@ -155,11 +184,25 @@ TEST_P(Test_Darknet_nets, YOLOv3) ...@@ -155,11 +184,25 @@ TEST_P(Test_Darknet_nets, YOLOv3)
classIds[0] = 7; confidences[0] = 0.952983f; boxes[0] = Rect2d(0.614622, 0.150257, 0.286747, 0.138994); // a truck classIds[0] = 7; confidences[0] = 0.952983f; boxes[0] = Rect2d(0.614622, 0.150257, 0.286747, 0.138994); // a truck
classIds[1] = 1; confidences[1] = 0.987908f; boxes[1] = Rect2d(0.150913, 0.221933, 0.591342, 0.524327); // a bycicle classIds[1] = 1; confidences[1] = 0.987908f; boxes[1] = Rect2d(0.150913, 0.221933, 0.591342, 0.524327); // a bycicle
classIds[2] = 16; confidences[2] = 0.998836f; boxes[2] = Rect2d(0.160024, 0.389964, 0.257861, 0.553752); // a dog (COCO) classIds[2] = 16; confidences[2] = 0.998836f; boxes[2] = Rect2d(0.160024, 0.389964, 0.257861, 0.553752); // a dog (COCO)
double scoreDiff = (targetId == DNN_TARGET_OPENCL_FP16 || targetId == DNN_TARGET_MYRIAD) ? 4e-3 : 8e-5;
double iouDiff = (targetId == DNN_TARGET_OPENCL_FP16 || targetId == DNN_TARGET_MYRIAD) ? 0.011 : 3e-5;
testDarknetModel("yolov3.cfg", "yolov3.weights", outNames, testDarknetModel("yolov3.cfg", "yolov3.weights", outNames,
classIds, confidences, boxes, targetId); classIds, confidences, boxes, backendId, targetId, scoreDiff, iouDiff);
} }
INSTANTIATE_TEST_CASE_P(/**/, Test_Darknet_nets, availableDnnTargets()); const tuple<DNNBackend, DNNTarget> testCases[] = {
#ifdef HAVE_INF_ENGINE
tuple<DNNBackend, DNNTarget>(DNN_BACKEND_INFERENCE_ENGINE, DNN_TARGET_CPU),
tuple<DNNBackend, DNNTarget>(DNN_BACKEND_INFERENCE_ENGINE, DNN_TARGET_OPENCL),
tuple<DNNBackend, DNNTarget>(DNN_BACKEND_INFERENCE_ENGINE, DNN_TARGET_OPENCL_FP16),
tuple<DNNBackend, DNNTarget>(DNN_BACKEND_INFERENCE_ENGINE, DNN_TARGET_MYRIAD),
#endif
tuple<DNNBackend, DNNTarget>(DNN_BACKEND_DEFAULT, DNN_TARGET_CPU),
tuple<DNNBackend, DNNTarget>(DNN_BACKEND_DEFAULT, DNN_TARGET_OPENCL),
tuple<DNNBackend, DNNTarget>(DNN_BACKEND_DEFAULT, DNN_TARGET_OPENCL_FP16)
};
INSTANTIATE_TEST_CASE_P(/**/, Test_Darknet_nets, testing::ValuesIn(testCases));
static void testDarknetLayer(const std::string& name, bool hasWeights = false) static void testDarknetLayer(const std::string& name, bool hasWeights = false)
{ {
......
...@@ -53,7 +53,7 @@ namespace opencv_test { ...@@ -53,7 +53,7 @@ namespace opencv_test {
using namespace cv::dnn; using namespace cv::dnn;
CV_ENUM(DNNBackend, DNN_BACKEND_DEFAULT, DNN_BACKEND_HALIDE, DNN_BACKEND_INFERENCE_ENGINE) CV_ENUM(DNNBackend, DNN_BACKEND_DEFAULT, DNN_BACKEND_HALIDE, DNN_BACKEND_INFERENCE_ENGINE)
CV_ENUM(DNNTarget, DNN_TARGET_CPU, DNN_TARGET_OPENCL, DNN_TARGET_OPENCL_FP16) CV_ENUM(DNNTarget, DNN_TARGET_CPU, DNN_TARGET_OPENCL, DNN_TARGET_OPENCL_FP16, DNN_TARGET_MYRIAD)
static testing::internal::ParamGenerator<DNNTarget> availableDnnTargets() static testing::internal::ParamGenerator<DNNTarget> availableDnnTargets()
{ {
......
...@@ -23,7 +23,7 @@ const char* keys = ...@@ -23,7 +23,7 @@ const char* keys =
"{ backend | 0 | Choose one of computation backends: " "{ backend | 0 | Choose one of computation backends: "
"0: default C++ backend, " "0: default C++ backend, "
"1: Halide language (http://halide-lang.org/), " "1: Halide language (http://halide-lang.org/), "
"2: Intel's Deep Learning Inference Engine (https://software.seek.intel.com/deep-learning-deployment)}" "2: Intel's Deep Learning Inference Engine (https://software.intel.com/openvino-toolkit)}"
"{ target | 0 | Choose one of target computation devices: " "{ target | 0 | Choose one of target computation devices: "
"0: CPU target (by default)," "0: CPU target (by default),"
"1: OpenCL }"; "1: OpenCL }";
......
...@@ -34,7 +34,7 @@ parser.add_argument('--backend', choices=backends, default=cv.dnn.DNN_BACKEND_DE ...@@ -34,7 +34,7 @@ parser.add_argument('--backend', choices=backends, default=cv.dnn.DNN_BACKEND_DE
help="Choose one of computation backends: " help="Choose one of computation backends: "
"%d: default C++ backend, " "%d: default C++ backend, "
"%d: Halide language (http://halide-lang.org/), " "%d: Halide language (http://halide-lang.org/), "
"%d: Intel's Deep Learning Inference Engine (https://software.seek.intel.com/deep-learning-deployment)" % backends) "%d: Intel's Deep Learning Inference Engine (https://software.intel.com/openvino-toolkit)" % backends)
parser.add_argument('--target', choices=targets, default=cv.dnn.DNN_TARGET_CPU, type=int, parser.add_argument('--target', choices=targets, default=cv.dnn.DNN_TARGET_CPU, type=int,
help='Choose one of target computation devices: ' help='Choose one of target computation devices: '
'%d: CPU target (by default), ' '%d: CPU target (by default), '
......
...@@ -25,7 +25,7 @@ const char* keys = ...@@ -25,7 +25,7 @@ const char* keys =
"{ backend | 0 | Choose one of computation backends: " "{ backend | 0 | Choose one of computation backends: "
"0: default C++ backend, " "0: default C++ backend, "
"1: Halide language (http://halide-lang.org/), " "1: Halide language (http://halide-lang.org/), "
"2: Intel's Deep Learning Inference Engine (https://software.seek.intel.com/deep-learning-deployment)}" "2: Intel's Deep Learning Inference Engine (https://software.intel.com/openvino-toolkit)}"
"{ target | 0 | Choose one of target computation devices: " "{ target | 0 | Choose one of target computation devices: "
"0: CPU target (by default)," "0: CPU target (by default),"
"1: OpenCL }"; "1: OpenCL }";
......
...@@ -35,7 +35,7 @@ parser.add_argument('--backend', choices=backends, default=cv.dnn.DNN_BACKEND_DE ...@@ -35,7 +35,7 @@ parser.add_argument('--backend', choices=backends, default=cv.dnn.DNN_BACKEND_DE
help="Choose one of computation backends: " help="Choose one of computation backends: "
"%d: default C++ backend, " "%d: default C++ backend, "
"%d: Halide language (http://halide-lang.org/), " "%d: Halide language (http://halide-lang.org/), "
"%d: Intel's Deep Learning Inference Engine (https://software.seek.intel.com/deep-learning-deployment)" % backends) "%d: Intel's Deep Learning Inference Engine (https://software.intel.com/openvino-toolkit)" % backends)
parser.add_argument('--target', choices=targets, default=cv.dnn.DNN_TARGET_CPU, type=int, parser.add_argument('--target', choices=targets, default=cv.dnn.DNN_TARGET_CPU, type=int,
help='Choose one of target computation devices: ' help='Choose one of target computation devices: '
'%d: CPU target (by default), ' '%d: CPU target (by default), '
......
...@@ -26,7 +26,7 @@ const char* keys = ...@@ -26,7 +26,7 @@ const char* keys =
"{ backend | 0 | Choose one of computation backends: " "{ backend | 0 | Choose one of computation backends: "
"0: default C++ backend, " "0: default C++ backend, "
"1: Halide language (http://halide-lang.org/), " "1: Halide language (http://halide-lang.org/), "
"2: Intel's Deep Learning Inference Engine (https://software.seek.intel.com/deep-learning-deployment)}" "2: Intel's Deep Learning Inference Engine (https://software.intel.com/openvino-toolkit)}"
"{ target | 0 | Choose one of target computation devices: " "{ target | 0 | Choose one of target computation devices: "
"0: CPU target (by default)," "0: CPU target (by default),"
"1: OpenCL }"; "1: OpenCL }";
......
...@@ -36,7 +36,7 @@ parser.add_argument('--backend', choices=backends, default=cv.dnn.DNN_BACKEND_DE ...@@ -36,7 +36,7 @@ parser.add_argument('--backend', choices=backends, default=cv.dnn.DNN_BACKEND_DE
help="Choose one of computation backends: " help="Choose one of computation backends: "
"%d: default C++ backend, " "%d: default C++ backend, "
"%d: Halide language (http://halide-lang.org/), " "%d: Halide language (http://halide-lang.org/), "
"%d: Intel's Deep Learning Inference Engine (https://software.seek.intel.com/deep-learning-deployment)" % backends) "%d: Intel's Deep Learning Inference Engine (https://software.intel.com/openvino-toolkit)" % backends)
parser.add_argument('--target', choices=targets, default=cv.dnn.DNN_TARGET_CPU, type=int, parser.add_argument('--target', choices=targets, default=cv.dnn.DNN_TARGET_CPU, type=int,
help='Choose one of target computation devices: ' help='Choose one of target computation devices: '
'%d: CPU target (by default), ' '%d: CPU target (by default), '
......
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