dnn: abstract dnn inference engine class and derive object detection class

      * clone openvino inference engine
          git clone https://github.com/opencv/dldt.git

      * build openvino inference engien
          https://github.com/opencv/dldt/blob/2018/inference-engine/README.md

      * export openvino path to environment varibles
          $ export OPENVINO_IE_INC_PATH=${OPENVINO_INFERENCE_ENGINE_DIR}
          $ export OPENVINO_IE_LIBS_PATH=${OPENVINO_INFERENCE_ENGINE_DIR}/bin/intel64/Release/lib

        # NOTE: Fill in the ${OPENVINO_INFERENCE_ENGINE_DIR} - e.g. /home/usr/openvino/dldt/inference-engine/

      * test command
          $ ./test-dnn-inference --input cars_on_road.jpg --model-file vehicle-detection-adas-0002.xml

8 files changed, 384 insertions, 156 deletions

diff --git a/configure.ac b/configure.ac
index 0922366..6c8c363 100644
--- a/configure.ac
+++ b/configure.ac
@@ -179,12 +179,12 @@ fi
 
 ENABLE_DNN=0
 if test "$enable_dnn" = "yes"; then
-    if test -z $OPENVINO_IE_CXXFLAGS; then
-        AC_MSG_WARN(Please export OPENVINO_IE_CXXFLAGS environment variables)
+    if test -z $OPENVINO_IE_INC_PATH; then
+        AC_MSG_WARN(Please export OPENVINO_IE_INC_PATH environment variables)
         AC_MSG_ERROR(OpenVino inc path has not been set ... disable DNN module!)
     fi
-    if test -z $OPENVINO_IE_LIBS; then
-        AC_MSG_WARN(Please export OPENVINO_IE_LIBS environment variables)
+    if test -z $OPENVINO_IE_LIBS_PATH; then
+        AC_MSG_WARN(Please export OPENVINO_IE_LIBS_PATH environment variables)
         AC_MSG_ERROR(OpenVino library path has not been set ... disable DNN module!)
     fi
     ENABLE_DNN=1
diff --git a/modules/dnn/Makefile.am b/modules/dnn/Makefile.am
index b75bc5d..bcad5fd 100644
--- a/modules/dnn/Makefile.am
+++ b/modules/dnn/Makefile.am
@@ -4,12 +4,14 @@ XCAM_DNN_CXXFLAGS = \
     $(XCAM_CXXFLAGS)        \
     -I$(top_srcdir)/xcore   \
     -I$(top_srcdir)/modules \
-    $(OPENVINO_IE_CXXFLAGS) \
+    -I$(OPENVINO_IE_INC_PATH)/include \
+    -I$(OPENVINO_IE_INC_PATH)/src/extension \
+    -I$(OPENVINO_IE_INC_PATH)/samples/common/format_reader \
     $(NULL)
 
 XCAM_DNN_LIBS = \
     $(top_builddir)/xcore/libxcam_core.la \
-    -L$(OPENVINO_IE_LIBS) \
+    -L$(OPENVINO_IE_LIBS_PATH) \
     -linference_engine \
     -lclDNN64 \
     -lclDNNPlugin \
@@ -20,6 +22,7 @@ XCAM_DNN_LIBS = \
 
 xcam_dnn_sources =       \
     dnn_inference_engine.cpp   \
+    dnn_object_detection.cpp   \
     $(NULL)
 
 libxcam_dnn_la_SOURCES = \
@@ -36,13 +39,14 @@ libxcam_dnn_la_LIBADD = \
 
 libxcam_dnn_la_LDFLAGS = \
     $(XCAM_LT_LDFLAGS)   \
-    -Wl,-rpath,$(OPENVINO_IE_LIBS) \
+    -Wl,-rpath,$(OPENVINO_IE_LIBS_PATH) \
     $(NULL)
 
 libxcam_dnnincludedir = $(includedir)/xcam/dnn
 
 nobase_libxcam_dnninclude_HEADERS = \
     dnn_inference_engine.h          \
+    dnn_object_detection.h          \
     $(NULL)
 
 libxcam_dnn_la_LIBTOOLFLAGS = --tag=disable-static
diff --git a/modules/dnn/dnn_inference_engine.cpp b/modules/dnn/dnn_inference_engine.cpp
index 256b2d8..8453161 100755
--- a/modules/dnn/dnn_inference_engine.cpp
+++ b/modules/dnn/dnn_inference_engine.cpp
@@ -18,7 +18,11 @@
  * Author: Zong Wei <wei.zong@intel.com>
  */
 
+#include <iomanip>
+
 #include <format_reader_ptr.h>
+#include <ext_list.hpp>
+
 #include "dnn_inference_engine.h"
 
 using namespace std;
@@ -62,10 +66,8 @@ DnnInferenceEngine::create_model (DnnInferConfig& config)
     _network_reader.ReadNetwork (get_filename_prefix (config.model_filename) + ".xml");
     _network_reader.ReadWeights (get_filename_prefix (config.model_filename) + ".bin");
 
-    // 2. Prepare inputs and outputs format
+    // 2. read network from model
     _network = _network_reader.getNetwork ();
-    _inputs_info = _network.getInputsInfo ();
-    _outputs_info = _network.getOutputsInfo ();
 
     // 3. Select Plugin - Select the plugin on which to load your network.
     // 3.1. Create the plugin with the InferenceEngine::PluginDispatcher load helper class.
@@ -241,13 +243,14 @@ DnnInferenceEngine::set_input_presion (uint32_t idx, DnnInferPrecisionType preci
     }
 
     uint32_t id = 0;
+    InputsDataMap inputs_info (_network.getInputsInfo ());
 
-    if (idx > _inputs_info.size ()) {
+    if (idx > inputs_info.size ()) {
         XCAM_LOG_ERROR ("Input is out of range");
         return XCAM_RETURN_ERROR_PARAM;
     }
 
-    for (auto & item : _inputs_info) {
+    for (auto & item : inputs_info) {
         if (id == idx) {
             Precision input_precision = convert_precision_type (precision);
             item.second->setPrecision (input_precision);
@@ -268,13 +271,14 @@ DnnInferenceEngine::get_input_presion (uint32_t idx)
     }
 
     uint32_t id = 0;
+    InputsDataMap inputs_info (_network.getInputsInfo ());
 
-    if (idx > _inputs_info.size ()) {
+    if (idx > inputs_info.size ()) {
         XCAM_LOG_ERROR ("Input is out of range");
         return DnnInferPrecisionUnspecified;
     }
-    /** Iterating over all input blobs **/
-    for (auto & item : _inputs_info) {
+
+    for (auto & item : inputs_info) {
         if (id == idx) {
             Precision input_precision = item.second->getPrecision ();
             return convert_precision_type (input_precision);
@@ -293,13 +297,14 @@ DnnInferenceEngine::set_output_presion (uint32_t idx, DnnInferPrecisionType prec
     }
 
     uint32_t id = 0;
+    OutputsDataMap outputs_info (_network.getOutputsInfo ());
 
-    if (idx > _outputs_info.size ()) {
+    if (idx > outputs_info.size ()) {
         XCAM_LOG_ERROR ("Output is out of range");
         return XCAM_RETURN_ERROR_PARAM;
     }
 
-    for (auto & item : _outputs_info) {
+    for (auto & item : outputs_info) {
         if (id == idx) {
             Precision output_precision = convert_precision_type (precision);
             item.second->setPrecision (output_precision);
@@ -320,13 +325,14 @@ DnnInferenceEngine::get_output_presion (uint32_t idx)
     }
 
     uint32_t id = 0;
+    OutputsDataMap outputs_info (_network.getOutputsInfo ());
 
-    if (idx > _outputs_info.size ()) {
+    if (idx > outputs_info.size ()) {
         XCAM_LOG_ERROR ("Input is out of range");
         return DnnInferPrecisionUnspecified;
     }
-    /** Iterating over all output blobs **/
-    for (auto & item : _outputs_info) {
+
+    for (auto & item : outputs_info) {
         if (id == idx) {
             Precision output_precision = item.second->getPrecision ();
             return convert_precision_type (output_precision);
@@ -344,13 +350,14 @@ DnnInferenceEngine::set_input_layout (uint32_t idx, DnnInferLayoutType layout)
         return XCAM_RETURN_ERROR_ORDER;
     }
     uint32_t id = 0;
+    InputsDataMap inputs_info (_network.getInputsInfo ());
 
-    if (idx > _inputs_info.size ()) {
+    if (idx > inputs_info.size ()) {
         XCAM_LOG_ERROR ("Input is out of range");
         return XCAM_RETURN_ERROR_PARAM;
     }
-    /** Iterating over all input blobs **/
-    for (auto & item : _inputs_info) {
+
+    for (auto & item : inputs_info) {
         if (id == idx) {
             /** Creating first input blob **/
             Layout input_layout = convert_layout_type (layout);
@@ -372,13 +379,14 @@ DnnInferenceEngine::set_output_layout (uint32_t idx, DnnInferLayoutType layout)
     }
 
     uint32_t id = 0;
+    OutputsDataMap outputs_info (_network.getOutputsInfo ());
 
-    if (idx > _outputs_info.size ()) {
+    if (idx > outputs_info.size ()) {
         XCAM_LOG_ERROR ("Output is out of range");
         return XCAM_RETURN_ERROR_PARAM;
     }
-    /** Iterating over all output blobs **/
-    for (auto & item : _outputs_info) {
+
+    for (auto & item : outputs_info) {
         if (id == idx) {
             Layout output_layout = convert_layout_type (layout);
             item.second->setLayout (output_layout);
@@ -391,36 +399,6 @@ DnnInferenceEngine::set_output_layout (uint32_t idx, DnnInferLayoutType layout)
 }
 
 XCamReturn
-DnnInferenceEngine::get_model_input_info (DnnInferInputOutputInfo& info)
-{
-    if (!_model_created) {
-        XCAM_LOG_ERROR ("Please create the model firstly!");
-        return XCAM_RETURN_ERROR_ORDER;
-    }
-
-    int id = 0;
-    for (auto & item : _inputs_info) {
-        auto& input = item.second;
-        const InferenceEngine::SizeVector input_dims = input->getDims ();
-
-        info.width[id] = input_dims[0];
-        info.height[id] = input_dims[1];
-        info.channels[id] = input_dims[2];
-        info.object_size[id] = input_dims[3];
-        info.precision[id] = convert_precision_type (input->getPrecision());
-        info.layout[id] = convert_layout_type (input->getLayout());
-
-        item.second->setPrecision(Precision::U8);
-
-        id++;
-    }
-    info.batch_size = get_batch_size ();
-    info.numbers = _inputs_info.size ();
-
-    return XCAM_RETURN_NO_ERROR;
-}
-
-XCamReturn
 DnnInferenceEngine::set_model_input_info (DnnInferInputOutputInfo& info)
 {
     XCAM_LOG_DEBUG ("DnnInferenceEngine::set_model_input_info");
@@ -430,13 +408,15 @@ DnnInferenceEngine::set_model_input_info (DnnInferInputOutputInfo& info)
         return XCAM_RETURN_ERROR_ORDER;
     }
 
-    if (info.numbers != _inputs_info.size ()) {
+    InputsDataMap inputs_info (_network.getInputsInfo ());
+
+    if (info.numbers != inputs_info.size ()) {
         XCAM_LOG_ERROR ("Input size is not matched with model info numbers %d !", info.numbers);
         return XCAM_RETURN_ERROR_PARAM;
     }
     int id = 0;
 
-    for (auto & item : _inputs_info) {
+    for (auto & item : inputs_info) {
         Precision precision = convert_precision_type (info.precision[id]);
         item.second->setPrecision (precision);
         Layout layout = convert_layout_type (info.layout[id]);
@@ -448,43 +428,6 @@ DnnInferenceEngine::set_model_input_info (DnnInferInputOutputInfo& info)
 }
 
 XCamReturn
-DnnInferenceEngine::get_model_output_info (DnnInferInputOutputInfo& info)
-{
-    if (!_model_created) {
-        XCAM_LOG_ERROR ("Please create the model firstly!");
-        return XCAM_RETURN_ERROR_ORDER;
-    }
-
-    int id = 0;
-    std::string output_name;
-    DataPtr output_info;
-    for (const auto& out : _outputs_info) {
-        if (out.second->creatorLayer.lock()->type == "DetectionOutput") {
-            output_name = out.first;
-            output_info = out.second;
-        }
-    }
-    if (output_info.get ()) {
-        const InferenceEngine::SizeVector output_dims = output_info->getTensorDesc().getDims();
-
-        info.width[id]    = output_dims[0];
-        info.height[id]   = output_dims[1];
-        info.channels[id] = output_dims[2];
-        info.object_size[id] = output_dims[3];
-
-        info.precision[id] = convert_precision_type (output_info->getPrecision());
-        info.layout[id] = convert_layout_type (output_info->getLayout());
-
-        info.batch_size = 1;
-        info.numbers = _outputs_info.size ();
-    } else {
-        XCAM_LOG_ERROR ("Get output info error!");
-        return XCAM_RETURN_ERROR_UNKNOWN;
-    }
-    return XCAM_RETURN_NO_ERROR;
-}
-
-XCamReturn
 DnnInferenceEngine::set_model_output_info (DnnInferInputOutputInfo& info)
 {
     if (!_model_created) {
@@ -492,13 +435,15 @@ DnnInferenceEngine::set_model_output_info (DnnInferInputOutputInfo& info)
         return XCAM_RETURN_ERROR_ORDER;
     }
 
-    if (info.numbers != _outputs_info.size()) {
+    OutputsDataMap outputs_info (_network.getOutputsInfo ());
+
+    if (info.numbers != outputs_info.size()) {
         XCAM_LOG_ERROR ("Output size is not matched with model!");
         return XCAM_RETURN_ERROR_PARAM;
     }
 
     int id = 0;
-    for (auto & item : _outputs_info) {
+    for (auto & item : outputs_info) {
         Precision precision = convert_precision_type (info.precision[id]);
         item.second->setPrecision (precision);
         Layout layout = convert_layout_type (info.layout[id]);
@@ -514,13 +459,14 @@ DnnInferenceEngine::set_input_blob (uint32_t idx, DnnInferData& data)
 {
     unsigned int id = 0;
     std::string item_name;
+    InputsDataMap inputs_info (_network.getInputsInfo ());
 
-    if (idx > _inputs_info.size()) {
+    if (idx > inputs_info.size()) {
         XCAM_LOG_ERROR ("Input is out of range");
         return XCAM_RETURN_ERROR_PARAM;
     }
 
-    for (auto & item : _inputs_info) {
+    for (auto & item : inputs_info) {
         if (id == idx) {
             item_name = item.first;
             break;
@@ -565,6 +511,8 @@ DnnInferenceEngine::set_inference_data (std::vector<std::string> images)
     }
 
     uint32_t idx = 0;
+    InputsDataMap inputs_info (_network.getInputsInfo ());
+
     for (auto & i : images) {
         FormatReader::ReaderPtr reader (i.c_str ());
         if (reader.get () == NULL) {
@@ -578,9 +526,9 @@ DnnInferenceEngine::set_inference_data (std::vector<std::string> images)
         uint32_t image_width = 0;
         uint32_t image_height = 0;
 
-        for (auto & item : _inputs_info) {
-            image_width = _inputs_info[item.first]->getDims()[0];
-            image_height = _inputs_info[item.first]->getDims()[1];
+        for (auto & item : inputs_info) {
+            image_width = inputs_info[item.first]->getDims()[0];
+            image_height = inputs_info[item.first]->getDims()[1];
         }
 
         std::shared_ptr<unsigned char> data (reader->getData (image_width, image_height));
@@ -634,42 +582,6 @@ DnnInferenceEngine::read_inference_image (std::string image)
     }
 }
 
-void*
-DnnInferenceEngine::get_inference_results (uint32_t idx, uint32_t& size)
-{
-    if (! _model_created || ! _model_loaded) {
-        XCAM_LOG_ERROR ("Please create and load the model firstly!");
-        return NULL;
-    }
-    uint32_t id = 0;
-    std::string item_name;
-
-    if (idx > _outputs_info.size ()) {
-        XCAM_LOG_ERROR ("Output is out of range");
-        return NULL;
-    }
-
-    for (auto & item : _outputs_info) {
-        if (item.second->creatorLayer.lock()->type == "DetectionOutput") {
-            item_name = item.first;
-            break;
-        }
-        id++;
-    }
-
-    if (item_name.empty ()) {
-        XCAM_LOG_ERROR ("item name is empty!");
-        return NULL;
-    }
-
-    const Blob::Ptr blob = _infer_request.GetBlob (item_name);
-    float* output_result = static_cast<PrecisionTrait<Precision::FP32>::value_type*>(blob->buffer ());
-
-    size = blob->byteSize ();
-
-    return (reinterpret_cast<void *>(output_result));
-}
-
 InferenceEngine::TargetDevice
 DnnInferenceEngine::get_device_from_string (const std::string &device_name)
 {
diff --git a/modules/dnn/dnn_inference_engine.h b/modules/dnn/dnn_inference_engine.h
index eebf93d..ed8ed36 100755
--- a/modules/dnn/dnn_inference_engine.h
+++ b/modules/dnn/dnn_inference_engine.h
@@ -25,10 +25,7 @@
 
 #include <vector>
 #include <string>
-#include <fstream>
-#include <iomanip>
 #include <inference_engine.hpp>
-#include <ext_list.hpp>
 
 #include <xcam_std.h>
 
@@ -229,14 +226,14 @@ public:
         return _input_image_width;
     };
 
-    XCamReturn set_model_input_info (DnnInferInputOutputInfo& info);
-    XCamReturn get_model_input_info (DnnInferInputOutputInfo& info);
+    virtual XCamReturn set_model_input_info (DnnInferInputOutputInfo& info) = 0;
+    virtual XCamReturn get_model_input_info (DnnInferInputOutputInfo& info) = 0;
 
-    XCamReturn set_model_output_info (DnnInferInputOutputInfo& info);
-    XCamReturn get_model_output_info (DnnInferInputOutputInfo& info);
+    virtual XCamReturn set_model_output_info (DnnInferInputOutputInfo& info) = 0;
+    virtual XCamReturn get_model_output_info (DnnInferInputOutputInfo& info) = 0;
 
-    XCamReturn set_inference_data (std::vector<std::string> images);
-    void* get_inference_results (uint32_t idx, uint32_t& size);
+    virtual XCamReturn set_inference_data (std::vector<std::string> images) = 0;
+    virtual void* get_inference_results (uint32_t idx, uint32_t& size) = 0;
 
     std::shared_ptr<uint8_t> read_inference_image (std::string image);
     void print_log (uint32_t flag);
@@ -257,8 +254,9 @@ protected:
 
     void print_performance_counts (const std::map<std::string, InferenceEngine::InferenceEngineProfileInfo>& performance_map);
 
-private:
     XCamReturn set_input_blob (uint32_t idx, DnnInferData& data);
+
+private:
     template <typename T> XCamReturn copy_image_to_blob (const DnnInferData& data, InferenceEngine::Blob::Ptr& blob, int batch_index);
     template <typename T> XCamReturn copy_data_to_blob (const DnnInferData& data, InferenceEngine::Blob::Ptr& blob, int batch_index);
 
@@ -267,18 +265,13 @@ protected:
     bool _model_created;
     bool _model_loaded;
 
-    InferenceEngine::InferencePlugin _plugin;
-
-    InferenceEngine::InputsDataMap _inputs_info;
-    InferenceEngine::OutputsDataMap _outputs_info;
-
     uint32_t _input_image_width;
     uint32_t _input_image_height;
 
+    InferenceEngine::InferencePlugin _plugin;
     InferenceEngine::CNNNetReader _network_reader;
     InferenceEngine::CNNNetwork _network;
     InferenceEngine::InferRequest _infer_request;
-
     std::vector<InferenceEngine::CNNLayerPtr> _layers;
 };
 
diff --git a/modules/dnn/dnn_object_detection.cpp b/modules/dnn/dnn_object_detection.cpp
new file mode 100644
index 0000000..42bf29d
--- /dev/null
+++ b/modules/dnn/dnn_object_detection.cpp
@@ -0,0 +1,264 @@
+/*
+ * dnn_object_detection.cpp -  object detection
+ *
+ *  Copyright (c) 2019 Intel Corporation
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * Author: Zong Wei <wei.zong@intel.com>
+ */
+
+#include <format_reader_ptr.h>
+
+#include "dnn_object_detection.h"
+
+using namespace std;
+using namespace InferenceEngine;
+
+namespace XCam {
+
+DnnObjectDetection::DnnObjectDetection (DnnInferConfig& config)
+    : DnnInferenceEngine (config)
+{
+    XCAM_LOG_DEBUG ("DnnObjectDetection::DnnObjectDetection");
+
+    create_model (config);
+}
+
+
+DnnObjectDetection::~DnnObjectDetection ()
+{
+
+}
+
+XCamReturn
+DnnObjectDetection::get_model_input_info (DnnInferInputOutputInfo& info)
+{
+    if (!_model_created) {
+        XCAM_LOG_ERROR ("Please create the model firstly!");
+        return XCAM_RETURN_ERROR_ORDER;
+    }
+
+    int id = 0;
+    InputsDataMap inputs_info (_network.getInputsInfo ());
+
+    for (auto & item : inputs_info) {
+        auto& input = item.second;
+        const InferenceEngine::SizeVector input_dims = input->getDims ();
+
+        info.width[id] = input_dims[0];
+        info.height[id] = input_dims[1];
+        info.channels[id] = input_dims[2];
+        info.object_size[id] = input_dims[3];
+        info.precision[id] = convert_precision_type (input->getPrecision());
+        info.layout[id] = convert_layout_type (input->getLayout());
+
+        item.second->setPrecision(Precision::U8);
+
+        id++;
+    }
+    info.batch_size = get_batch_size ();
+    info.numbers = inputs_info.size ();
+
+    return XCAM_RETURN_NO_ERROR;
+}
+
+XCamReturn
+DnnObjectDetection::set_model_input_info (DnnInferInputOutputInfo& info)
+{
+    XCAM_LOG_DEBUG ("DnnObjectDetection::set_model_input_info");
+
+    if (!_model_created) {
+        XCAM_LOG_ERROR ("Please create the model firstly!");
+        return XCAM_RETURN_ERROR_ORDER;
+    }
+
+    InputsDataMap inputs_info (_network.getInputsInfo ());
+    if (info.numbers != inputs_info.size ()) {
+        XCAM_LOG_ERROR ("Input size is not matched with model info numbers %d !", info.numbers);
+        return XCAM_RETURN_ERROR_PARAM;
+    }
+    int id = 0;
+
+    for (auto & item : inputs_info) {
+        Precision precision = convert_precision_type (info.precision[id]);
+        item.second->setPrecision (precision);
+        Layout layout = convert_layout_type (info.layout[id]);
+        item.second->setLayout (layout);
+        id++;
+    }
+
+    return XCAM_RETURN_NO_ERROR;
+}
+
+XCamReturn
+DnnObjectDetection::get_model_output_info (DnnInferInputOutputInfo& info)
+{
+    if (!_model_created) {
+        XCAM_LOG_ERROR ("Please create the model firstly!");
+        return XCAM_RETURN_ERROR_ORDER;
+    }
+
+    int id = 0;
+    std::string output_name;
+    OutputsDataMap outputs_info (_network.getOutputsInfo ());
+    DataPtr output_info;
+    for (const auto& out : outputs_info) {
+        if (out.second->creatorLayer.lock()->type == "DetectionOutput") {
+            output_name = out.first;
+            output_info = out.second;
+            break;
+        }
+    }
+    if (output_info.get ()) {
+        const InferenceEngine::SizeVector output_dims = output_info->getTensorDesc().getDims();
+
+        info.width[id]    = output_dims[0];
+        info.height[id]   = output_dims[1];
+        info.channels[id] = output_dims[2];
+        info.object_size[id] = output_dims[3];
+
+        info.precision[id] = convert_precision_type (output_info->getPrecision());
+        info.layout[id] = convert_layout_type (output_info->getLayout());
+
+        info.batch_size = 1;
+        info.numbers = outputs_info.size ();
+    } else {
+        XCAM_LOG_ERROR ("Get output info error!");
+        return XCAM_RETURN_ERROR_UNKNOWN;
+    }
+    return XCAM_RETURN_NO_ERROR;
+}
+
+XCamReturn
+DnnObjectDetection::set_model_output_info (DnnInferInputOutputInfo& info)
+{
+    if (!_model_created) {
+        XCAM_LOG_ERROR ("Please create the model firstly!");
+        return XCAM_RETURN_ERROR_ORDER;
+    }
+
+    OutputsDataMap outputs_info (_network.getOutputsInfo ());
+    if (info.numbers != outputs_info.size ()) {
+        XCAM_LOG_ERROR ("Output size is not matched with model!");
+        return XCAM_RETURN_ERROR_PARAM;
+    }
+
+    int id = 0;
+    for (auto & item : outputs_info) {
+        Precision precision = convert_precision_type (info.precision[id]);
+        item.second->setPrecision (precision);
+        Layout layout = convert_layout_type (info.layout[id]);
+        item.second->setLayout (layout);
+        id++;
+    }
+
+    return XCAM_RETURN_NO_ERROR;
+}
+
+XCamReturn
+DnnObjectDetection::set_inference_data (std::vector<std::string> images)
+{
+    if (!_model_created) {
+        XCAM_LOG_ERROR ("Please create the model firstly!");
+        return XCAM_RETURN_ERROR_ORDER;
+    }
+
+    uint32_t idx = 0;
+    InputsDataMap inputs_info (_network.getInputsInfo ());
+
+    for (auto & i : images) {
+        FormatReader::ReaderPtr reader (i.c_str ());
+        if (reader.get () == NULL) {
+            XCAM_LOG_WARNING ("Image %d cannot be read!", i);
+            continue;
+        }
+
+        _input_image_width = reader->width ();
+        _input_image_height = reader->height ();
+
+        uint32_t image_width = 0;
+        uint32_t image_height = 0;
+
+        for (auto & item : inputs_info) {
+            image_width = inputs_info[item.first]->getDims ()[0];
+            image_height = inputs_info[item.first]->getDims ()[1];
+        }
+
+        std::shared_ptr<unsigned char> data (reader->getData (image_width, image_height));
+
+        if (data.get () != NULL) {
+            DnnInferData image;
+            image.width = image_width;
+            image.height = image_height;
+            image.width_stride = image_width;
+            image.height_stride = image_height;
+            image.buffer = data.get ();
+            image.channel_num = 3;
+            image.batch_idx = idx;
+            image.image_format = DnnInferImageFormatBGRPacked;
+
+            // set precision & data type
+            image.precision = get_input_presion (idx);
+            image.data_type = DnnInferDataTypeImage;
+
+            set_input_blob (idx, image);
+
+            idx ++;
+        } else {
+            XCAM_LOG_WARNING ("Valid input images were not found!");
+            continue;
+        }
+    }
+
+    return XCAM_RETURN_NO_ERROR;
+}
+
+void*
+DnnObjectDetection::get_inference_results (uint32_t idx, uint32_t& size)
+{
+    if (! _model_created || ! _model_loaded) {
+        XCAM_LOG_ERROR ("Please create and load the model firstly!");
+        return NULL;
+    }
+    uint32_t id = 0;
+    std::string item_name;
+
+    OutputsDataMap outputs_info (_network.getOutputsInfo ());
+    if (idx > outputs_info.size ()) {
+        XCAM_LOG_ERROR ("Output is out of range");
+        return NULL;
+    }
+
+    for (auto & item : outputs_info) {
+        if (item.second->creatorLayer.lock()->type == "DetectionOutput") {
+            item_name = item.first;
+            break;
+        }
+        id++;
+    }
+
+    if (item_name.empty ()) {
+        XCAM_LOG_ERROR ("item name is empty!");
+        return NULL;
+    }
+
+    const Blob::Ptr blob = _infer_request.GetBlob (item_name);
+    float* output_result = static_cast<PrecisionTrait<Precision::FP32>::value_type*>(blob->buffer ());
+
+    size = blob->byteSize ();
+
+    return (reinterpret_cast<void *>(output_result));
+}
+
+}  // namespace XCam
diff --git a/modules/dnn/dnn_object_detection.h b/modules/dnn/dnn_object_detection.h
new file mode 100644
index 0000000..6980858
--- /dev/null
+++ b/modules/dnn/dnn_object_detection.h
@@ -0,0 +1,53 @@
+/*
+ * dnn_object_detection.h -  object detection
+ *
+ *  Copyright (c) 2019 Intel Corporation
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ * Author: Zong Wei <wei.zong@intel.com>
+ */
+
+#ifndef _XCMA_DNN_OBJECT_DETECTION_H_
+#define _XCMA_DNN_OBJECT_DETECTION_H_
+
+#pragma once
+
+#include <string>
+#include <inference_engine.hpp>
+
+#include <xcam_std.h>
+#include "dnn_inference_engine.h"
+
+namespace XCam {
+
+class DnnObjectDetection
+    : public DnnInferenceEngine
+{
+public:
+    explicit DnnObjectDetection (DnnInferConfig& config);
+    virtual ~DnnObjectDetection ();
+
+    XCamReturn set_model_input_info (DnnInferInputOutputInfo& info);
+    XCamReturn get_model_input_info (DnnInferInputOutputInfo& info);
+
+    XCamReturn set_model_output_info (DnnInferInputOutputInfo& info);
+    XCamReturn get_model_output_info (DnnInferInputOutputInfo& info);
+
+    XCamReturn set_inference_data (std::vector<std::string> images);
+    void* get_inference_results (uint32_t idx, uint32_t& size);
+};
+
+}  // namespace XCam
+
+#endif //_XCMA_DNN_INFERENCE_ENGINE_H_
diff --git a/tests/Makefile.am b/tests/Makefile.am
index d16c72e..eb27a9e 100644
--- a/tests/Makefile.am
+++ b/tests/Makefile.am
@@ -241,7 +241,8 @@ TEST_DNN_INFERENCE_LA = $(top_builddir)/modules/dnn/libxcam_dnn.la
 test_dnn_inference_SOURCES = test-dnn-inference.cpp
 test_dnn_inference_CXXFLAGS = \
     $(TEST_BASE_CXXFLAGS) \
-    $(OPENVINO_IE_CXXFLAGS) \
+    -I$(OPENVINO_IE_INC_PATH)/include \
+    -I$(OPENVINO_IE_INC_PATH)/src/extension \
     $(NULL)
 
 test_dnn_inference_LDADD = \
diff --git a/tests/test-dnn-inference.cpp b/tests/test-dnn-inference.cpp
index f8f7662..d07e27c 100644
--- a/tests/test-dnn-inference.cpp
+++ b/tests/test-dnn-inference.cpp
@@ -29,6 +29,7 @@
 #include "test_common.h"
 
 #include <dnn/dnn_inference_engine.h>
+#include <dnn/dnn_object_detection.h>
 
 using namespace XCam;
 using namespace InferenceEngine;
@@ -346,7 +347,7 @@ int main (int argc, char *argv[])
     XCAM_LOG_DEBUG ("2. Create inference engine");
     infer_config.perf_counter = 0;
 
-    SmartPtr<DnnInferenceEngine> infer_engine = new DnnInferenceEngine (infer_config);
+    SmartPtr<DnnInferenceEngine> infer_engine = new DnnObjectDetection (infer_config);
 
     DnnInferenceEngineInfo infer_info;
     CHECK (