Remove ResultBase and clean up label

Author: RHeckerIntel

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

@@ -17,33 +17,16 @@
 
 struct MetaData;
 
-struct ResultBase {
-    ResultBase(int64_t frameId = -1, const std::shared_ptr<MetaData>& metaData = nullptr)
+struct InferenceResult {
+    InferenceResult(int64_t frameId = -1, const std::shared_ptr<MetaData>& metaData = nullptr)
         : frameId(frameId),
           metaData(metaData) {}
-    virtual ~ResultBase() {}
 
+    std::shared_ptr<InternalModelData> internalModelData;
+    std::map<std::string, ov::Tensor> outputsData;
     int64_t frameId;
-
     std::shared_ptr<MetaData> metaData;
-    bool IsEmpty() {
-        return frameId < 0;
-    }
-
-    template <class T>
-    T& asRef() {
-        return dynamic_cast<T&>(*this);
-    }
-
-    template <class T>
-    const T& asRef() const {
-        return dynamic_cast<const T&>(*this);
-    }
-};
 
-struct InferenceResult : public ResultBase {
-    std::shared_ptr<InternalModelData> internalModelData;
-    std::map<std::string, ov::Tensor> outputsData;
 
     /// Returns the first output tensor
     /// This function is a useful addition to direct access to outputs list as many models have only one output
@@ -89,35 +72,22 @@ struct DetectedKeypoints {
 class Label {
 public:
     Label() {}
-    Label(int id, std::string name):  id(id), name(name) {}
+    Label(int id, std::string name, float score): id(id), name(name), score(score) {}
 
     int id;
     std::string name;
-
-    friend std::ostream& operator<< (std::ostream& os, const Label& label) {
-        return os << label.id << " (" << label.name << ")";
-    }
-};
-
-class LabelScore {
-public:
-    LabelScore() {}
-    LabelScore(int id, std::string name, float score): label(Label(id, name)), score(score) {}
-    LabelScore(Label label, float score):  label(label), score(score) {}
-
-    Label label;
     float score;
 
-    friend std::ostream& operator<< (std::ostream& os, const LabelScore& label) {
-        return os << label.label << ": " << std::fixed << std::setprecision(3) << label.score;
+    friend std::ostream& operator<< (std::ostream& os, const Label& label) {
+        return os << label.id << " (" << label.name << ")" << ": " << std::fixed << std::setprecision(3) << label.score;
     }
 };
 
 class Mask {
 public:
-    Mask(LabelScore label, cv::Rect roi, cv::Mat mask): label(label), roi(roi), mask(mask) {}
+    Mask(Label label, cv::Rect roi, cv::Mat mask): label(label), roi(roi), mask(mask) {}
 
-    LabelScore label;
+    Label label;
     cv::Rect roi;
     cv::Mat mask;
 
@@ -152,16 +122,16 @@ static inline std::vector<Contour> getContours(const std::vector<Mask>& segmente
         if (contours.size() != 1) {
             throw std::runtime_error("findContours() must have returned only one contour");
         }
-        combined_contours.push_back({obj.label.label.name, obj.label.score, contours[0]});
+        combined_contours.push_back({obj.label.name, obj.label.score, contours[0]});
     }
     return combined_contours;
 }
 
 class Box {
 public:
-    Box(cv::Rect shape, std::vector<LabelScore> labels): shape(shape), labels(labels) {}
+    Box(cv::Rect shape, std::vector<Label> labels): shape(shape), labels(labels) {}
     cv::Rect shape;
-    std::vector<LabelScore> labels;
+    std::vector<Label> labels;
 
     friend std::ostream& operator<< (std::ostream& os, const Box& box) {
 
@@ -189,7 +159,7 @@ class Box {
 
 class RotatedRect {
 public:
-    LabelScore label;
+    Label label;
     cv::RotatedRect shape;
 
     friend std::ostream& operator<< (std::ostream& os, const RotatedRect& box) {

src/cpp/models/src/anomaly_model.cpp

@@ -86,7 +86,7 @@ std::unique_ptr<Scene> AnomalyModel::postprocess(InferenceResult& infResult) {
     auto scene = std::make_unique<Scene>(infResult.frameId, infResult.metaData);
 
     scene->saliency_maps.push_back(anomaly_map);
-    auto label = LabelScore(label_id, pred_label, pred_score);
+    auto label = Label(label_id, pred_label, pred_score);
     auto roi = cv::Rect(0, 0, inputImgSize.inputImgWidth, inputImgSize.inputImgHeight);
     scene->masks.push_back(Mask(label, roi, pred_mask));
     scene->boxes.push_back(Box(roi, {label}));
@@ -100,7 +100,7 @@ std::unique_ptr<Scene> AnomalyModel::postprocess(InferenceResult& infResult) {
             scene->boxes.push_back(
                 Box(
                     rect,
-                    {LabelScore(label_id, pred_label, box_score)}
+                    {Label(label_id, pred_label, box_score)}
                 )
             );
 

src/cpp/models/src/classification_model.cpp

@@ -328,7 +328,7 @@ std::unique_ptr<Scene> ClassificationModel::get_multilabel_predictions(Inference
 
     auto scene = std::make_unique<Scene>(infResult.frameId, infResult.metaData);
     auto raw_scores = ov::Tensor();
-    std::vector<LabelScore> result;
+    std::vector<Label> result;
     float* raw_scoresPtr = nullptr;
     if (add_raw_scores) {
         raw_scores = ov::Tensor(logitsTensor.get_element_type(), logitsTensor.get_shape());
@@ -401,9 +401,9 @@ std::unique_ptr<Scene> ClassificationModel::get_hierarchical_predictions(Inferen
     }
 
     auto resolved_labels = resolver->resolve_labels(predicted_labels, predicted_scores);
-    std::vector<LabelScore> result;
+    std::vector<Label> result;
     for (const auto& label : resolved_labels) {
-        result.push_back(LabelScore(hierarchical_info.label_to_idx[label.first], label.first, label.second));
+        result.push_back(Label(hierarchical_info.label_to_idx[label.first], label.first, label.second));
     }
     const auto& internalData = infResult.internalModelData->asRef<InternalImageModelData>();
     cv::Rect shape(0, 0, internalData.inputImgWidth, internalData.inputImgHeight);
@@ -449,7 +449,7 @@ std::unique_ptr<Scene> ClassificationModel::get_multiclass_predictions(Inference
         scene->additional_tensors["raw_scores"] = raw_scores;
     }
 
-    std::vector<LabelScore> result;
+    std::vector<Label> result;
     for (size_t i = 0; i < scoresTensor.get_size(); ++i) {
         int ind = indicesPtr[i];
         if (ind < 0 || ind >= static_cast<int>(labels.size())) {

src/cpp/models/src/detection_model_ssd.cpp

@@ -163,7 +163,7 @@ std::unique_ptr<Scene> ModelSSD::postprocessSingleOutput(InferenceResult& infRes
                         0.f,
                         floatInputImgHeight) - y
                 ),
-                {LabelScore(labelID, getLabelName(labelID), confidence)}
+                {Label(labelID, getLabelName(labelID), confidence)}
             );
             scene->boxes.push_back(box);
         }
@@ -223,7 +223,7 @@ std::unique_ptr<Scene> ModelSSD::postprocessMultipleOutputs(InferenceResult& inf
             if (width * height >= box_area_threshold) {
                 scene->boxes.push_back(Box(
                   cv::Rect(x, y, width, height),
-                  {LabelScore(labels[i], getLabelName(labels[i]), confidence)}
+                  {Label(labels[i], getLabelName(labels[i]), confidence)}
                 ));
             }
         }

src/cpp/models/src/detection_model_yolo.cpp

@@ -282,7 +282,7 @@ std::unique_ptr<Scene> ModelYolo::postprocess(InferenceResult& infResult) {
         for (const auto& obj1 : objects) {
             bool isGoodResult = true;
             for (const auto& obj2 : objects) {
-                if (obj1.labels[0].label.id == obj2.labels[0].label.id && obj1.labels[0].score < obj2.labels[0].score &&
+                if (obj1.labels[0].id == obj2.labels[0].id && obj1.labels[0].score < obj2.labels[0].score &&
                     intersectionOverUnion(obj1, obj2) >= iou_threshold) {  // if obj1 is the same as obj2, condition
                                                                            // expression will evaluate to false anyway
                     isGoodResult = false;
@@ -410,7 +410,7 @@ void ModelYolo::parseYOLOOutput(const std::string& output_name,
 
                     //--- Checking confidence threshold conformance and adding region to the list
                     if (prob >= confidence_threshold) {
-                        objects.push_back(Box(obj, {LabelScore(j, getLabelName(j), prob)}));
+                        objects.push_back(Box(obj, {Label(j, getLabelName(j), prob)}));
                     }
                 }
             }
@@ -635,7 +635,7 @@ std::unique_ptr<Scene> YOLOv5::postprocess(InferenceResult& infResult) {
 
         scene->boxes.push_back(Box(
             cv::Rect(x, y, width, height),
-            {LabelScore(labelID, label, confidence)}
+            {Label(labelID, label, confidence)}
         ));
     }
 

src/cpp/models/src/detection_model_yolov3_onnx.cpp

@@ -163,7 +163,7 @@ std::unique_ptr<Scene> ModelYoloV3ONNX::postprocess(InferenceResult& infResult)
             obj.height = clamp(height, 0.f, static_cast<float>(imgHeight));
             obj.width = clamp(width, 0.f, static_cast<float>(imgWidth));
 
-            scene->boxes.push_back(Box(obj, {LabelScore(classInd, getLabelName(classInd), score)}));
+            scene->boxes.push_back(Box(obj, {Label(classInd, getLabelName(classInd), score)}));
         }
     }
     return scene;

src/cpp/models/src/detection_model_yolox.cpp

@@ -195,7 +195,7 @@ std::unique_ptr<Scene> ModelYoloX::postprocess(InferenceResult& infResult) {
         obj.width =
             clamp(validBoxes[index].right - validBoxes[index].left, 0.f, static_cast<float>(scale.inputImgWidth));
         scene->boxes.push_back(
-            Box(obj, {LabelScore(classes[index], getLabelName(classes[index]), scores[index])})
+            Box(obj, {Label(classes[index], getLabelName(classes[index]), scores[index])})
         );
     }
     return scene;

src/cpp/models/src/instance_segmentation.cpp

@@ -8,8 +8,6 @@
 #include <stddef.h>
 #include <stdint.h>
 
-#include <fstream>
-#include <limits>
 #include <opencv2/core.hpp>
 #include <opencv2/imgproc.hpp>
 #include <openvino/openvino.hpp>
@@ -322,7 +320,7 @@ std::unique_ptr<Scene> MaskRCNNModel::postprocess(InferenceResult& infResult) {
             continue;
         }
 
-        LabelScore label(labelID, getLabelName(labelID), confidence);
+        Label label(labelID, getLabelName(labelID), confidence);
 
         cv::Rect roi;
 

src/cpp/models/src/segmentation_model.cpp

@@ -261,7 +261,7 @@ std::unique_ptr<Scene> SegmentationModel::postprocess(InferenceResult& infResult
 
     auto scene = std::make_unique<Scene>(infResult.frameId, infResult.metaData);
     auto roi = cv::Rect(0, 0, inputImgSize.inputImgWidth, inputImgSize.inputImgHeight);
-    scene->masks.push_back(Mask(LabelScore(0, "hard_prediction", 0), roi, hard_prediction));
+    scene->masks.push_back(Mask(Label(0, "hard_prediction", 0), roi, hard_prediction));
     if (return_soft_prediction) {
         cv::resize(soft_prediction,
                    soft_prediction,
@@ -270,7 +270,7 @@ std::unique_ptr<Scene> SegmentationModel::postprocess(InferenceResult& infResult
                    0.0,
                    cv::INTER_NEAREST);
 
-        scene->masks.push_back(Mask(LabelScore(1, "soft_prediction", 0), roi, soft_prediction));
+        scene->masks.push_back(Mask(Label(1, "soft_prediction", 0), roi, soft_prediction));
         auto iter = infResult.outputsData.find(feature_vector_name);
         if (infResult.outputsData.end() != iter) {
             scene->saliency_maps.push_back(get_activation_map(soft_prediction));

src/cpp/tilers/src/detection.cpp

@@ -66,7 +66,7 @@ std::unique_ptr<Scene> DetectionTiler::merge_results(const std::vector<std::uniq
 
     for (const auto& result : tiles_results) {
         for (auto& det : result->boxes) {
-            all_detections.emplace_back(det.shape.x, det.shape.y, det.shape.x + det.shape.width, det.shape.y + det.shape.height, det.labels[0].label.id);
+            all_detections.emplace_back(det.shape.x, det.shape.y, det.shape.x + det.shape.width, det.shape.y + det.shape.height, det.labels[0].id);
             all_scores.push_back(det.labels[0].score);
             all_detections_refs.push_back(det);
         }