Text Detection model DB (#175)

* Text Detection model DB

* review 1

models/text_detection_db/CMakeLists.txt

@@ -0,0 +1,29 @@
+cmake_minimum_required(VERSION 3.24)
+set(project_name "opencv_zoo_text_detection_db")
+
+PROJECT (${project_name})
+
+set(OPENCV_VERSION "4.7.0")
+set(OPENCV_INSTALLATION_PATH "" CACHE PATH "Where to look for OpenCV installation")
+find_package(OpenCV ${OPENCV_VERSION} REQUIRED HINTS ${OPENCV_INSTALLATION_PATH})
+# Find OpenCV, you may need to set OpenCV_DIR variable
+# to the absolute path to the directory containing OpenCVConfig.cmake file
+# via the command line or GUI
+
+file(GLOB SourceFile
+    "demo.cpp")
+# If the package has been found, several variables will
+# be set, you can find the full list with descriptions
+# in the OpenCVConfig.cmake file.
+# Print some message showing some of them
+message(STATUS "OpenCV library status:")
+message(STATUS "    config: ${OpenCV_DIR}")
+message(STATUS "    version: ${OpenCV_VERSION}")
+message(STATUS "    libraries: ${OpenCV_LIBS}")
+message(STATUS "    include path: ${OpenCV_INCLUDE_DIRS}")
+
+# Declare the executable target built from your sources
+add_executable(${project_name} ${SourceFile})
+
+# Link your application with OpenCV libraries
+target_link_libraries(${project_name} PRIVATE ${OpenCV_LIBS})

models/text_detection_db/README.md

@@ -11,6 +11,8 @@ Note:
 
 ## Demo
 
+### Python
+
 Run the following command to try the demo:
 
 ```shell
@@ -23,6 +25,22 @@ python demo.py --input /path/to/image -v
 python demo.py --help
 ```
 
+### C++
+
+Install latest OpenCV and CMake >= 3.24.0 to get started with:
+
+```shell
+# A typical and default installation path of OpenCV is /usr/local
+cmake -B build -D OPENCV_INSTALLATION_PATH=/path/to/opencv/installation .
+cmake --build build
+# detect on camera input
+./build/opencv_zoo_text_detection_db -m=/path/to/model
+# detect on an image
+./build/opencv_zoo_text_detection_db -m=/path/to/model -i=/path/to/image -v
+# get help messages
+./build/opencv_zoo_text_detection_db -h
+```
+
 ### Example outputs
 
 ![mask](./example_outputs/mask.jpg)

models/text_detection_db/demo.cpp

@@ -0,0 +1,179 @@
+#include <iostream>
+
+#include <opencv2/dnn.hpp>
+#include <opencv2/imgproc.hpp>
+#include <opencv2/highgui.hpp>
+
+using namespace std;
+using namespace cv;
+using namespace dnn;
+
+vector< pair<cv::dnn::Backend, cv::dnn::Target> > backendTargetPairs = {
+        std::make_pair<cv::dnn::Backend, cv::dnn::Target>(dnn::DNN_BACKEND_OPENCV, dnn::DNN_TARGET_CPU),
+        std::make_pair<cv::dnn::Backend, cv::dnn::Target>(dnn::DNN_BACKEND_CUDA, dnn::DNN_TARGET_CUDA),
+        std::make_pair<cv::dnn::Backend, cv::dnn::Target>(dnn::DNN_BACKEND_CUDA, dnn::DNN_TARGET_CUDA_FP16),
+        std::make_pair<cv::dnn::Backend, cv::dnn::Target>(dnn::DNN_BACKEND_TIMVX, dnn::DNN_TARGET_NPU),
+        std::make_pair<cv::dnn::Backend, cv::dnn::Target>(dnn::DNN_BACKEND_CANN, dnn::DNN_TARGET_NPU)};
+        
+
+std::string keys =
+"{ help  h           |                                              | Print help message. }"
+"{ model m           | text_detection_DB_IC15_resnet18_2021sep.onnx | Usage: Set model type, defaults to text_detection_DB_IC15_resnet18_2021sep.onnx }"
+"{ input i           |                                              | Usage: Path to input image or video file. Skip this argument to capture frames from a camera.}"
+"{ width             | 736                                          | Usage: Resize input image to certain width, default = 736. It should be multiple by 32.}"
+"{ height            | 736                                          | Usage: Resize input image to certain height, default = 736. It should be multiple by 32.}"
+"{ binary_threshold  | 0.3                                          | Usage: Threshold of the binary map, default = 0.3.}"
+"{ polygon_threshold | 0.5                                          | Usage: Threshold of polygons, default = 0.5.}"
+"{ max_candidates    | 200                                          | Usage: Set maximum number of polygon candidates, default = 200.}"
+"{ unclip_ratio      | 2.0                                          | Usage: The unclip ratio of the detected text region, which determines the output size, default = 2.0.}"
+"{ save s            | true                                         | Usage: Specify to save file with results (i.e. bounding box, confidence level). Invalid in case of camera input.}"
+"{ viz v             | true                                         | Usage: Specify to open a new window to show results. Invalid in case of camera input.}"
+"{ backend bt        | 0                                            | Choose one of computation backends: "
+"0: (default) OpenCV implementation + CPU, "
+"1: CUDA + GPU (CUDA), "
+"2: CUDA + GPU (CUDA FP16), "
+"3: TIM-VX + NPU, "
+"4: CANN + NPU}";
+
+
+class DB {
+public:
+
+    DB(string modPath, Size inSize = Size(736, 736), float binThresh = 0.3,
+        float polyThresh = 0.5, int maxCand = 200, double unRatio = 2.0,
+        dnn::Backend bId = DNN_BACKEND_DEFAULT, dnn::Target tId = DNN_TARGET_CPU) : modelPath(modPath), inputSize(inSize), binaryThreshold(binThresh),
+        polygonThreshold(polyThresh), maxCandidates(maxCand), unclipRatio(unRatio),
+        backendId(bId), targetId(tId)
+    {
+        this->model = TextDetectionModel_DB(readNet(modelPath));
+        this->model.setPreferableBackend(backendId);
+        this->model.setPreferableTarget(targetId);
+
+        this->model.setBinaryThreshold(binaryThreshold);
+        this->model.setPolygonThreshold(polygonThreshold);
+        this->model.setUnclipRatio(unclipRatio);
+        this->model.setMaxCandidates(maxCandidates);
+
+        this->model.setInputParams(1.0 / 255.0, inputSize, Scalar(122.67891434, 116.66876762, 104.00698793));
+    }
+    pair< vector<vector<Point>>, vector<float> > infer(Mat image) {
+        CV_Assert(image.rows == this->inputSize.height && "height of input image != net input size ");
+        CV_Assert(image.cols == this->inputSize.width && "width of input image != net input size ");
+        vector<vector<Point>> pt;
+        vector<float> confidence;
+        this->model.detect(image, pt, confidence);
+        return make_pair< vector<vector<Point>> &, vector< float > &>(pt, confidence);
+    }
+
+private:
+    string modelPath;
+    TextDetectionModel_DB model;
+    Size inputSize;
+    float binaryThreshold;
+    float polygonThreshold;
+    int maxCandidates;
+    double unclipRatio;
+    dnn::Backend backendId;
+    dnn::Target targetId;
+
+};
+
+Mat visualize(Mat image, pair< vector<vector<Point>>, vector<float> >&results, double fps=-1, Scalar boxColor=Scalar(0, 255, 0), Scalar textColor=Scalar(0, 0, 255), bool isClosed=true, int thickness=2)
+{
+    Mat output;
+    image.copyTo(output);
+    if (fps > 0)
+        putText(output, format("FPS: %.2f", fps), Point(0, 15), FONT_HERSHEY_SIMPLEX, 0.5, textColor);
+    polylines(output, results.first, isClosed, boxColor, thickness);
+    return output;
+}
+
+int main(int argc, char** argv)
+{
+    CommandLineParser parser(argc, argv, keys);
+
+    parser.about("Use this program to run Real-time Scene Text Detection with Differentiable Binarization in opencv Zoo  using OpenCV.");
+    if (parser.has("help"))
+    {
+        parser.printMessage();
+        return 0;
+    }
+
+    int backendTargetid = parser.get<int>("backend");
+    String modelName = parser.get<String>("model");
+
+    if (modelName.empty())
+    {
+        CV_Error(Error::StsError, "Model file " + modelName + " not found");
+    }
+
+    Size inpSize(parser.get<int>("width"), parser.get<int>("height"));
+    float binThresh = parser.get<float>("binary_threshold");
+    float polyThresh = parser.get<float>("polygon_threshold");
+    int maxCand = parser.get<int>("max_candidates");
+    double unRatio = parser.get<float>("unclip_ratio");
+    bool save = parser.get<bool>("save");
+    bool viz = parser.get<float>("viz");
+
+    DB model(modelName, inpSize, binThresh, polyThresh, maxCand, unRatio, backendTargetPairs[backendTargetid].first, backendTargetPairs[backendTargetid].second);
+
+    //! [Open a video file or an image file or a camera stream]
+    VideoCapture cap;
+    if (parser.has("input"))
+        cap.open(parser.get<String>("input"));
+    else
+        cap.open(0);
+    if (!cap.isOpened())
+        CV_Error(Error::StsError, "Cannot opend video or file");
+    Mat originalImage;
+    static const std::string kWinName = modelName;
+    while (waitKey(1) < 0)
+    {
+        cap >> originalImage;
+        if (originalImage.empty())
+        {
+            cout << "Frame is empty" << endl;
+            waitKey();
+            break;
+        }
+        int originalW = originalImage.cols;
+        int originalH = originalImage.rows;
+        double scaleHeight = originalH / double(inpSize.height);
+        double scaleWidth = originalW / double(inpSize.width);
+        Mat image;
+        resize(originalImage, image, inpSize);
+
+        // inference
+        TickMeter tm;
+        tm.start();
+        pair< vector<vector<Point>>, vector<float> > results = model.infer(image);
+        tm.stop();
+        auto x = results.first;
+        // Scale the results bounding box
+        for (auto &pts : results.first)
+        {
+            for (int i = 0; i < 4; i++)
+            {
+                pts[i].x = int(pts[i].x * scaleWidth);
+                pts[i].y = int(pts[i].y * scaleHeight);
+            }
+        }
+        originalImage = visualize(originalImage, results, tm.getFPS());
+        tm.reset();
+        if (parser.has("input"))
+        {
+            if (save)
+            {
+                cout << "Result image saved to result.jpg\n";
+                imwrite("result.jpg", originalImage);
+            }
+            if (viz)
+                imshow(kWinName, originalImage);
+        }
+        else
+            imshow(kWinName, originalImage);
+    }
+    return 0;
+}
+
+