Reverted change for gaze estimation, added logs for gesture recognition, added check for s-t model

Author: Pashchenkov Maxim

demos/common/cpp/models/src/style_transfer_model.cpp

@@ -40,7 +40,12 @@ void StyleTransferModel::prepareInputsOutputs(InferenceEngine::CNNNetwork& cnnNe
     SizeVector& inSizeVector = inputShapes.begin()->second;
     if (inSizeVector.size() != 4 || inSizeVector[0] != 1 || inSizeVector[1] != 3)
         throw std::runtime_error("3-channel 4-dimensional model's input is expected");
-    InputInfo& inputInfo = *cnnNetwork.getInputsInfo().begin()->second;
+    auto inputsDataMap = cnnNetwork.getInputsInfo();
+    if (inputsDataMap.empty()) {
+        throw std::runtime_error("Network input data node information is empty");
+    }
+    InputInfo& inputInfo = *inputsDataMap.begin()->second;
+
     inputInfo.setPrecision(Precision::FP32);
 
     // --------------------------- Prepare output blobs -----------------------------------------------------

demos/gaze_estimation_demo/cpp/src/landmarks_estimator.cpp

@@ -98,7 +98,10 @@ std::vector<cv::Point2i> LandmarksEstimator::heatMapPostprocess(cv::Rect faceBou
 
 std::vector<cv::Mat> LandmarksEstimator::split(std::vector<float>& data, const std::vector<unsigned long>& shape) {
     std::vector<cv::Mat> flattenData(shape[1]);
-    float* output = data.data();
+    float* output = new float[data.size()];
+    for (size_t i = 0; i < data.size(); i++) {
+        output[i] = data[i];
+    }
     for (size_t i = 0; i < flattenData.size(); i++) {
         flattenData[i] = cv::Mat(shape[2], shape[3], CV_32FC1, output + i * shape[2] * shape[3]);
     }

demos/gesture_recognition_demo/cpp_gapi/main.cpp

@@ -101,15 +101,19 @@ int main(int argc, char *argv[]) {
             FLAGS_m_d,                         // path to model
             fileNameNoExt(FLAGS_m_d) + ".bin", // path to weights
             FLAGS_d_d                          // device to use
-        }.cfgOutputLayers({"boxes"}); // This clarification here because
-                                      // of GAPI take the last layer from .xml
-                                      // and last layer sould be the outpul layer
+        }.cfgOutputLayers({"boxes"}); // This clarification here because of
+                                      // GAPI take the first layer name from OutputsInfo
+                                      // for one output G_API_NET API
+        slog::info << "The Person Detection ASL model " << FLAGS_m_d << " is loaded to " << FLAGS_d_d << " device." << slog::endl;
 
         auto action_recognition = cv::gapi::ie::Params<nets::ActionRecognition> {
             FLAGS_m_a,                         // path to model
             fileNameNoExt(FLAGS_m_a) + ".bin", // path to weights
             FLAGS_d_a                          // device to use
-        }.cfgOutputLayers({"output"}); // The same
+        }.cfgOutputLayers({"output"}); // This clarification here because of
+                                       // GAPI take the first layer name from OutputsInfo
+                                       // for one output G_API_NET API
+        slog::info << "The Action Recognition model " << FLAGS_m_a << " is loaded to " << FLAGS_d_a << " device." << slog::endl;
 
         /** Custom kernels **/
         auto kernels = custom::kernels();