Preview of DetectionResult refactoring

examples/cpp/CMakeLists.txt

@@ -92,7 +92,7 @@ endmacro()
 
 find_package(OpenCV REQUIRED COMPONENTS imgcodecs)
 
-set (ENABLE_PY_BINDINGS OFF)
+set(ENABLE_PY_BINDINGS OFF)
 add_subdirectory(../../src/cpp ${Samples_BINARY_DIR}/src/cpp)
 
 add_example(NAME asynchronous_api SOURCES ./asynchronous_api/main.cpp DEPENDENCIES model_api)

examples/cpp/asynchronous_api/main.cpp

@@ -12,7 +12,7 @@
 #include <iomanip>
 #include <iostream>
 #include <opencv2/core.hpp>
-#include <opencv2/imgproc.hpp>
+#include <opencv2/imgcodecs.hpp>
 #include <openvino/openvino.hpp>
 #include <stdexcept>
 #include <string>

examples/cpp/synchronous_api/main.cpp

@@ -13,8 +13,7 @@
 #include <iomanip>
 #include <iostream>
 #include <opencv2/core.hpp>
-#include <opencv2/highgui.hpp>
-#include <opencv2/imgproc.hpp>
+#include <opencv2/imgcodecs.hpp>
 #include <openvino/openvino.hpp>
 #include <stdexcept>
 #include <string>
@@ -41,6 +40,12 @@ int main(int argc, char* argv[]) try {
                   << std::setw(4) << int(obj.x) << " | " << std::setw(4) << int(obj.y) << " | " << std::setw(4)
                   << int(obj.x + obj.width) << " | " << std::setw(4) << int(obj.y + obj.height) << "\n";
     }
+
+    // could be improved with an iterator
+    for (size_t i = 0; i < result->size(); ++i) {
+        std::cout << result->getDetection(i) << "\n";
+    }
+
 } catch (const std::exception& error) {
     std::cerr << error.what() << '\n';
     return 1;

src/cpp/models/include/models/results.h

@@ -147,6 +147,9 @@ struct ClassificationResult : public ResultBase {
 };
 
 struct DetectedObject : public cv::Rect2f {
+    DetectedObject() {}
+    DetectedObject(const cv::Rect2f& rect) : cv::Rect2f(rect) {}
+
     size_t labelID;
     std::string label;
     float confidence;
@@ -161,8 +164,27 @@ struct DetectedObject : public cv::Rect2f {
 struct DetectionResult : public ResultBase {
     DetectionResult(int64_t frameId = -1, const std::shared_ptr<MetaData>& metaData = nullptr)
         : ResultBase(frameId, metaData) {}
-    std::vector<DetectedObject> objects;
-    ov::Tensor saliency_map, feature_vector;  // Contan "saliency_map" and "feature_vector" model outputs if such exist
+
+    // memory to forward as np.array
+    cv::Mat_<float> scores;
+    cv::Mat_<float> bboxes;
+    cv::Mat_<int> labels;
+    std::vector<std::string> label_names;
+
+    std::vector<DetectedObject> objects; // to remove
+    ov::Tensor saliency_map, feature_vector;  // Contain "saliency_map" and "feature_vector" model outputs if such exist
+
+    size_t size() const {
+        return label_names.size();
+    }
+
+    DetectedObject getDetection(size_t idx) {
+        DetectedObject result(bboxes.at<cv::Rect2f>(idx));
+        result.labelID = labels.at<size_t>(idx);
+        result.label = label_names[idx];
+        result.confidence = scores.at<float>(idx);
+        return result;
+    }
 
     friend std::ostream& operator<<(std::ostream& os, const DetectionResult& prediction) {
         for (const DetectedObject& obj : prediction.objects) {

src/cpp/models/src/detection_model_ssd.cpp

@@ -131,6 +131,7 @@ std::unique_ptr<ResultBase> ModelSSD::postprocessSingleOutput(InferenceResult& i
         }
     }
 
+    result->label_names.reserve(numAndStep.detectionsNum);
     for (size_t i = 0; i < numAndStep.detectionsNum; i++) {
         float image_id = detections[i * numAndStep.objectSize + 0];
         if (image_id < 0) {
@@ -165,6 +166,12 @@ std::unique_ptr<ResultBase> ModelSSD::postprocessSingleOutput(InferenceResult& i
                       floatInputImgHeight) -
                 desc.y;
             result->objects.push_back(desc);
+
+            auto label_idx = static_cast<int>(detections[i * numAndStep.objectSize + 1]);
+            result->labels.at<int>(i) = label_idx;
+            result->scores.at<float>(i) = confidence;
+            result->bboxes.at<cv::Rect2f>(i) = cv::Rect2f(desc.x, desc.y, desc.width, desc.height);
+            result->label_names.push_back(getLabelName(label_idx));
         }
     }
 
@@ -200,6 +207,18 @@ std::unique_ptr<ResultBase> ModelSSD::postprocessMultipleOutputs(InferenceResult
     float widthScale = scores ? netInputWidth : 1.0f;
     float heightScale = scores ? netInputHeight : 1.0f;
 
+    size_t detections_num = 0;
+    for (size_t i = 0; i < numAndStep.detectionsNum; i++) {
+        float confidence = scores ? scores[i] : boxes[i * numAndStep.objectSize + 4];
+        if (confidence > confidence_threshold) {
+            ++detections_num;
+        }
+    }
+    result->label_names.reserve(detections_num);
+    result->labels = cv::Mat(1, detections_num, CV_32S);
+    result->scores = cv::Mat(1, detections_num, CV_32F);
+    result->bboxes = cv::Mat(1, detections_num, CV_32FC4);
+
     for (size_t i = 0; i < numAndStep.detectionsNum; i++) {
         float confidence = scores ? scores[i] : boxes[i * numAndStep.objectSize + 4];
 
@@ -229,6 +248,11 @@ std::unique_ptr<ResultBase> ModelSSD::postprocessMultipleOutputs(InferenceResult
             if (desc.width * desc.height >= box_area_threshold) {
                 result->objects.push_back(desc);
             }
+            auto label_idx = static_cast<int>(labels[i]);
+            result->labels.at<int>(i) = label_idx;
+            result->scores.at<float>(i) = confidence;
+            result->bboxes.at<cv::Rect2f>(i) = cv::Rect2f(desc.x, desc.y, desc.width, desc.height);
+            result->label_names.push_back(getLabelName(label_idx));
         }
     }