Further refactor

msluszniak · msluszniak · commit b3c5834ce56b · 2025-09-02T09:42:23.000+02:00
diff --git a/packages/react-native-executorch/common/rnexecutorch/data_processing/ImageProcessing.cpp b/packages/react-native-executorch/common/rnexecutorch/data_processing/ImageProcessing.cpp
@@ -175,7 +175,7 @@ cv::Mat convertTensorToMatrix(cv::Size size, const Tensor &tensor) {
       std::span<const float>(resultData, tensor.numel()), size);
 }
 
-static cv::Size getTensorSize(const std::vector<int32_t> &tensorDims) {
+cv::Size getSizeOfImageFromTensorDims(const std::vector<int32_t> &tensorDims) {
   if (tensorDims.size() < 2) {
     throw std::runtime_error(
         "Unexpected tensor size, expected at least 2 dimentions but got: " +
@@ -185,14 +185,11 @@ static cv::Size getTensorSize(const std::vector<int32_t> &tensorDims) {
   return {tensorDims[tensorDims.size() - 1], tensorDims[tensorDims.size() - 2]};
 }
 
-TensorPtr covertMatrixToTensor(const std::vector<int32_t> &tensorDims,
-                               cv::Mat &input) {
-
-  cv::Size tensorSize = getTensorSize(tensorDims);
+void adaptImageForTensor(const std::vector<int32_t> &tensorDims,
+                         cv::Mat &input) {
+  cv::Size tensorSize = getSizeOfImageFromTensorDims(tensorDims);
   cv::resize(input, input, tensorSize);
   cv::cvtColor(input, input, cv::COLOR_BGR2RGB);
-
-  return getTensorFromMatrix(tensorDims, input);
 }
 
 static cv::Scalar
diff --git a/packages/react-native-executorch/common/rnexecutorch/data_processing/ImageProcessing.h b/packages/react-native-executorch/common/rnexecutorch/data_processing/ImageProcessing.h
@@ -34,9 +34,11 @@ cv::Mat readImageToMatrix(const std::string &imageURI);
 /// @brief Create an OpenCV matrix based on a tensor content
 cv::Mat convertTensorToMatrix(cv::Size size, const Tensor &tensor);
 
-/// @brief Create a tensor based on an OpenCV matrix content. Please mind that
-/// for performance reasons matrix is passed by mutable reference
-TensorPtr covertMatrixToTensor(const std::vector<int32_t> &tensorDims,
-                               cv::Mat &input);
+cv::Size getSizeOfImageFromTensorDims(const std::vector<int32_t> &tensorDims);
+
+/// @brief Resize and convert color of an image passed as matrix
+/// to fit tensor requirements
+void adaptImageForTensor(const std::vector<int32_t> &tensorDims,
+                         cv::Mat &input);
 
 } // namespace rnexecutorch::image_processing
diff --git a/packages/react-native-executorch/common/rnexecutorch/models/classification/Classification.cpp b/packages/react-native-executorch/common/rnexecutorch/models/classification/Classification.cpp
@@ -16,23 +16,17 @@ Classification::Classification(const std::string &modelSource,
     throw std::runtime_error("Model seems to not take any input tensors.");
   }
   std::vector<int32_t> modelInputShape = inputShapes[0];
-  if (modelInputShape.size() < 2) {
-    char errorMessage[100];
-    std::snprintf(errorMessage, sizeof(errorMessage),
-                  "Unexpected model input size, expected at least 2 dimentions "
-                  "but got: %zu.",
-                  modelInputShape.size());
-    throw std::runtime_error(errorMessage);
-  }
-  modelImageSize = cv::Size(modelInputShape[modelInputShape.size() - 1],
-                            modelInputShape[modelInputShape.size() - 2]);
+  modelImageSize =
+      image_processing::getSizeOfImageFromTensorDims(modelInputShape);
 }
 
 std::unordered_map<std::string_view, float>
 Classification::generate(std::string imageSource) {
   auto imageAsMatrix = image_processing::readImageToMatrix(imageSource);
-  auto inputTensor = image_processing::covertMatrixToTensor(
-      getAllInputShapes()[0], imageAsMatrix);
+  const auto tensorDims = getAllInputShapes()[0];
+  image_processing::adaptImageForTensor(tensorDims, imageAsMatrix);
+  auto inputTensor =
+      image_processing::getTensorFromMatrix(tensorDims, imageAsMatrix);
   auto forwardResult = BaseModel::forward(inputTensor);
   if (!forwardResult.ok()) {
     throw std::runtime_error(
diff --git a/packages/react-native-executorch/common/rnexecutorch/models/embeddings/image/ImageEmbeddings.cpp b/packages/react-native-executorch/common/rnexecutorch/models/embeddings/image/ImageEmbeddings.cpp
@@ -15,23 +15,17 @@ ImageEmbeddings::ImageEmbeddings(
     throw std::runtime_error("Model seems to not take any input tensors.");
   }
   std::vector<int32_t> modelInputShape = inputTensors[0];
-  if (modelInputShape.size() < 2) {
-    char errorMessage[100];
-    std::snprintf(errorMessage, sizeof(errorMessage),
-                  "Unexpected model input size, expected at least 2 dimentions "
-                  "but got: %zu.",
-                  modelInputShape.size());
-    throw std::runtime_error(errorMessage);
-  }
-  modelImageSize = cv::Size(modelInputShape[modelInputShape.size() - 1],
-                            modelInputShape[modelInputShape.size() - 2]);
+  modelImageSize =
+      image_processing::getSizeOfImageFromTensorDims(modelInputShape);
 }
 
 std::shared_ptr<OwningArrayBuffer>
 ImageEmbeddings::generate(std::string imageSource) {
   auto imageAsMatrix = image_processing::readImageToMatrix(imageSource);
-  auto inputTensor = image_processing::covertMatrixToTensor(
-      getAllInputShapes()[0], imageAsMatrix);
+  const auto tensorDims = getAllInputShapes()[0];
+  image_processing::adaptImageForTensor(tensorDims, imageAsMatrix);
+  auto inputTensor =
+      image_processing::getTensorFromMatrix(tensorDims, imageAsMatrix);
 
   auto forwardResult = BaseModel::forward(inputTensor);
   if (!forwardResult.ok()) {
diff --git a/packages/react-native-executorch/common/rnexecutorch/models/image_segmentation/ImageSegmentation.cpp b/packages/react-native-executorch/common/rnexecutorch/models/image_segmentation/ImageSegmentation.cpp
@@ -20,16 +20,8 @@ ImageSegmentation::ImageSegmentation(
     throw std::runtime_error("Model seems to not take any input tensors.");
   }
   std::vector<int32_t> modelInputShape = inputShapes[0];
-  if (modelInputShape.size() < 2) {
-    char errorMessage[100];
-    std::snprintf(errorMessage, sizeof(errorMessage),
-                  "Unexpected model input size, expected at least 2 dimentions "
-                  "but got: %zu.",
-                  modelInputShape.size());
-    throw std::runtime_error(errorMessage);
-  }
-  modelImageSize = cv::Size(modelInputShape[modelInputShape.size() - 1],
-                            modelInputShape[modelInputShape.size() - 2]);
+  modelImageSize =
+      image_processing::getSizeOfImageFromTensorDims(modelInputShape);
   numModelPixels = modelImageSize.area();
 }
 
@@ -38,8 +30,10 @@ std::shared_ptr<jsi::Object> ImageSegmentation::generate(
     std::set<std::string, std::less<>> classesOfInterest, bool resize) {
   auto imageAsMatrix = image_processing::readImageToMatrix(imageSource);
   auto originalSize = imageAsMatrix.size();
-  auto inputTensor = image_processing::covertMatrixToTensor(
-      getAllInputShapes()[0], imageAsMatrix);
+  const auto tensorDims = getAllInputShapes()[0];
+  image_processing::adaptImageForTensor(tensorDims, imageAsMatrix);
+  auto inputTensor =
+      image_processing::getTensorFromMatrix(tensorDims, imageAsMatrix);
 
   auto forwardResult = BaseModel::forward(inputTensor);
   if (!forwardResult.ok()) {
diff --git a/packages/react-native-executorch/common/rnexecutorch/models/object_detection/ObjectDetection.cpp b/packages/react-native-executorch/common/rnexecutorch/models/object_detection/ObjectDetection.cpp
@@ -13,16 +13,8 @@ ObjectDetection::ObjectDetection(
     throw std::runtime_error("Model seems to not take any input tensors.");
   }
   std::vector<int32_t> modelInputShape = inputTensors[0];
-  if (modelInputShape.size() < 2) {
-    char errorMessage[100];
-    std::snprintf(errorMessage, sizeof(errorMessage),
-                  "Unexpected model input size, expected at least 2 dimentions "
-                  "but got: %zu.",
-                  modelInputShape.size());
-    throw std::runtime_error(errorMessage);
-  }
-  modelImageSize = cv::Size(modelInputShape[modelInputShape.size() - 1],
-                            modelInputShape[modelInputShape.size() - 2]);
+  modelImageSize =
+      image_processing::getSizeOfImageFromTensorDims(modelInputShape);
 }
 
 std::vector<types::Detection>
@@ -69,8 +61,10 @@ std::vector<types::Detection>
 ObjectDetection::generate(std::string imageSource, double detectionThreshold) {
   auto imageAsMatrix = image_processing::readImageToMatrix(imageSource);
   auto originalSize = imageAsMatrix.size();
-  auto inputTensor = image_processing::covertMatrixToTensor(
-      getAllInputShapes()[0], imageAsMatrix);
+  const auto tensorDims = getAllInputShapes()[0];
+  image_processing::adaptImageForTensor(tensorDims, imageAsMatrix);
+  auto inputTensor =
+      image_processing::getTensorFromMatrix(tensorDims, imageAsMatrix);
 
   auto forwardResult = BaseModel::forward(inputTensor);
   if (!forwardResult.ok()) {
diff --git a/packages/react-native-executorch/common/rnexecutorch/models/ocr/Detector.cpp b/packages/react-native-executorch/common/rnexecutorch/models/ocr/Detector.cpp
@@ -13,13 +13,8 @@ Detector::Detector(const std::string &modelSource,
         "Detector model seems to not take any input tensors.");
   }
   std::vector<int32_t> modelInputShape = inputShapes[0];
-  if (modelInputShape.size() < 2) {
-    throw std::runtime_error("Unexpected detector model input size, expected "
-                             "at least 2 dimensions but got: " +
-                             std::to_string(modelInputShape.size()) + ".");
-  }
-  modelImageSize = cv::Size(modelInputShape[modelInputShape.size() - 1],
-                            modelInputShape[modelInputShape.size() - 2]);
+  modelImageSize =
+      image_processing::getSizeOfImageFromTensorDims(modelInputShape);
 }
 
 cv::Size Detector::getModelImageSize() const noexcept { return modelImageSize; }
diff --git a/packages/react-native-executorch/common/rnexecutorch/models/ocr/Recognizer.cpp b/packages/react-native-executorch/common/rnexecutorch/models/ocr/Recognizer.cpp
@@ -15,11 +15,8 @@ Recognizer::Recognizer(const std::string &modelSource,
     throw std::runtime_error("Recognizer model has no input tensors.");
   }
   std::vector<int32_t> modelInputShape = inputShapes[0];
-  if (modelInputShape.size() < 2) {
-    throw std::runtime_error("Unexpected Recognizer model input shape.");
-  }
-  modelImageSize = cv::Size(modelInputShape[modelInputShape.size() - 1],
-                            modelInputShape[modelInputShape.size() - 2]);
+  modelImageSize =
+      image_processing::getSizeOfImageFromTensorDims(modelInputShape);
 }
 
 std::pair<std::vector<int32_t>, float>
diff --git a/packages/react-native-executorch/common/rnexecutorch/models/style_transfer/StyleTransfer.cpp b/packages/react-native-executorch/common/rnexecutorch/models/style_transfer/StyleTransfer.cpp
@@ -18,16 +18,8 @@ StyleTransfer::StyleTransfer(const std::string &modelSource,
     throw std::runtime_error("Model seems to not take any input tensors.");
   }
   std::vector<int32_t> modelInputShape = inputShapes[0];
-  if (modelInputShape.size() < 2) {
-    char errorMessage[100];
-    std::snprintf(errorMessage, sizeof(errorMessage),
-                  "Unexpected model input size, expected at least 2 dimentions "
-                  "but got: %zu.",
-                  modelInputShape.size());
-    throw std::runtime_error(errorMessage);
-  }
-  modelImageSize = cv::Size(modelInputShape[modelInputShape.size() - 1],
-                            modelInputShape[modelInputShape.size() - 2]);
+  modelImageSize =
+      image_processing::getSizeOfImageFromTensorDims(modelInputShape);
 }
 
 std::string StyleTransfer::postprocess(const Tensor &tensor,
@@ -41,8 +33,10 @@ std::string StyleTransfer::postprocess(const Tensor &tensor,
 std::string StyleTransfer::generate(std::string imageSource) {
   auto imageAsMatrix = image_processing::readImageToMatrix(imageSource);
   auto originalSize = imageAsMatrix.size();
-  auto inputTensor = image_processing::covertMatrixToTensor(
-      getAllInputShapes()[0], imageAsMatrix);
+  const auto tensorDims = getAllInputShapes()[0];
+  image_processing::adaptImageForTensor(tensorDims, imageAsMatrix);
+  auto inputTensor =
+      image_processing::getTensorFromMatrix(tensorDims, imageAsMatrix);
 
   auto forwardResult = BaseModel::forward(inputTensor);
   if (!forwardResult.ok()) {

Original file line number	Diff line number	Diff line change
`@@ -15,11 +15,8 @@ Recognizer::Recognizer(const std::string &modelSource,`
`15`	`15`	`throw std::runtime_error("Recognizer model has no input tensors.");`
`16`	`16`	`}`
`17`	`17`	`std::vector<int32_t> modelInputShape = inputShapes[0];`
`18`		`- if (modelInputShape.size() < 2) {`
`19`		`- throw std::runtime_error("Unexpected Recognizer model input shape.");`
`20`		`- }`
`21`		`- modelImageSize = cv::Size(modelInputShape[modelInputShape.size() - 1],`
`22`		`- modelInputShape[modelInputShape.size() - 2]);`
	`18`	`+ modelImageSize =`
	`19`	`+ image_processing::getSizeOfImageFromTensorDims(modelInputShape);`
`23`	`20`	`}`
`24`	`21`
`25`	`22`	`std::pair<std::vector<int32_t>, float>`