Preview of DetectionResult refactoring

.github/labeler.yml

@@ -3,12 +3,12 @@
 cpp:
   - changed-files:
       - any-glob-to-any-file:
-          - model_api/cpp/**
+          - src/cpp/**
 
 python:
   - changed-files:
       - any-glob-to-any-file:
-          - model_api/python/**
+          - src/python/**
 
 tests:
   - changed-files:

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>
@@ -36,7 +35,7 @@ int main(int argc, char* argv[]) try {
     auto result = model->infer(image);
 
     // Process detections
-    for (auto& obj : result->objects) {
+    for (auto obj : *result) {
         std::cout << " " << std::left << std::setw(9) << obj.label << " | " << std::setw(10) << obj.confidence << " | "
                   << 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";

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;
@@ -161,11 +164,55 @@ 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
+
+    cv::Mat_<float> scores;
+    cv::Mat_<float> bboxes;
+    cv::Mat_<int> labels;
+    std::vector<std::string> label_names;
+    ov::Tensor saliency_map, feature_vector;  // Contain "saliency_map" and "feature_vector" model outputs if such exist
+
+    struct DetectionIterator {
+        const DetectionResult& result;
+        size_t index;
+
+        DetectionIterator(const DetectionResult& result, size_t index) : result(result), index(index) {}
+
+        bool operator!=(const DetectionIterator& other) const {
+            return index != other.index;
+        }
+
+        DetectionIterator& operator++() {
+            ++index;
+            return *this;
+        }
+
+        DetectedObject operator*() const {
+            return result.getObject(index);
+        }
+    };
+
+    DetectionIterator begin() const {
+        return DetectionIterator(*this, 0);
+    }
+
+    DetectionIterator end() const {
+        return DetectionIterator(*this, size());
+    }
+
+    size_t size() const {
+        return label_names.size();
+    }
+
+    DetectedObject getObject(size_t idx) const {
+        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) {
+        for (const DetectedObject obj : prediction) {
             os << obj << "; ";
         }
         try {

src/cpp/models/src/detection_model_ssd.cpp

@@ -131,6 +131,18 @@ std::unique_ptr<ResultBase> ModelSSD::postprocessSingleOutput(InferenceResult& i
         }
     }
 
+    size_t detections_num = 0;
+    for (size_t i = 0; i < numAndStep.detectionsNum; i++) {
+        float confidence = detections[i * numAndStep.objectSize + 2];
+        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 image_id = detections[i * numAndStep.objectSize + 0];
         if (image_id < 0) {
@@ -141,11 +153,7 @@ std::unique_ptr<ResultBase> ModelSSD::postprocessSingleOutput(InferenceResult& i
 
         /** Filtering out objects with confidence < confidence_threshold probability **/
         if (confidence > confidence_threshold) {
-            DetectedObject desc;
-
-            desc.confidence = confidence;
-            desc.labelID = static_cast<size_t>(detections[i * numAndStep.objectSize + 1]);
-            desc.label = getLabelName(desc.labelID);
+            cv::Rect2f desc;
             desc.x =
                 clamp(round((detections[i * numAndStep.objectSize + 3] * netInputWidth - padLeft) * invertedScaleX),
                       0.f,
@@ -164,7 +172,13 @@ std::unique_ptr<ResultBase> ModelSSD::postprocessSingleOutput(InferenceResult& i
                       0.f,
                       floatInputImgHeight) -
                 desc.y;
-            result->objects.push_back(desc);
+
+            size_t det_idx = result->label_names.size();
+            auto label_idx = static_cast<int>(detections[i * numAndStep.objectSize + 1]);
+            result->labels.at<int>(det_idx) = label_idx;
+            result->scores.at<float>(det_idx) = confidence;
+            result->bboxes.at<cv::Rect2f>(det_idx) = desc;
+            result->label_names.push_back(getLabelName(label_idx));
         }
     }
 
@@ -200,16 +214,25 @@ 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];
 
         /** Filtering out objects with confidence < confidence_threshold probability **/
         if (confidence > confidence_threshold) {
-            DetectedObject desc;
+            cv::Rect2f desc;
 
-            desc.confidence = confidence;
-            desc.labelID = labels[i];
-            desc.label = getLabelName(desc.labelID);
             desc.x = clamp_and_round((boxes[i * numAndStep.objectSize] * widthScale - padLeft) * invertedScaleX,
                                      0.f,
                                      floatInputImgWidth);
@@ -229,6 +252,13 @@ std::unique_ptr<ResultBase> ModelSSD::postprocessMultipleOutputs(InferenceResult
             if (desc.width * desc.height >= box_area_threshold) {
                 result->objects.push_back(desc);
             }
+
+            size_t det_idx = result->label_names.size();
+            auto label_idx = static_cast<int>(labels[i]);
+            result->labels.at<int>(det_idx) = label_idx;
+            result->scores.at<float>(det_idx) = confidence;
+            result->bboxes.at<cv::Rect2f>(det_idx) = desc;
+            result->label_names.push_back(getLabelName(label_idx));
         }
     }