Face_detector.Cpp

Face detector using the BlazeFace Mediapipe model (with both CPU and TPU delegates) written in C++;

• Plain C/C++ implementation with minimal dependencies (Tensorflow Lite + OpenCV)
• Google MediaPipe models without the Mediapipe framework
• Support for Coral Edge TPU (not supported when using the ROS2 binding)
• Runs on ARM as well (Tested on RPI 3,4 and 5)
• ROS2 support

API Features

This library offers support for:

• Face detection
• Face ROI (Region of interest) detection
• 2D Landmarks of the Left eye, Right Eye, Nose tip, Mouth, Left eye tragion and Right eye tragion

Face detection

This is some example code for face detection:

    #define MODEL_PATH FACE_DETECTOR_MODEL_DIR "/CPU/face_detection.tflite"
    ...

    /* Initialize face detector library */
    CLFML::FaceDetection::FaceDetector det;
    
    /* Load model and initialize inference runtime */
    det.load_model(MODEL_PATH);
    
    /* Load image into model and do inference! */
    det.load_image(cam_frame);

    /* Get face_detected value */
    const int face_detected = det.detected(); // returns -1 for no face and 0 for face detected!

Region Of Interest

This is some example code for capturing the region of interest:

    #define MODEL_PATH FACE_DETECTOR_MODEL_DIR "/CPU/face_detection.tflite"
    ...

    /* Initialize face detector library */
    CLFML::FaceDetection::FaceDetector det;
    
    /* Load model and initialize inference runtime */
    det.load_model(MODEL_PATH);
    
    /* Load image into model and do inference! */
    det.load_image(cam_frame);

    /* Get the face roi rectangle */
    cv::Rect face_roi = det.get_face_roi();

Example code

For a full example showcasing both these API functions see the example code in example/face_roi_demo/demo.cpp.

Building with CMake

Before using this library you will need the following packages installed:

• OpenCV
• Working C++ compiler (GCC, Clang, MSVC (2017 or Higher))
• CMake
• Ninja (Optional, but preferred)

Running the examples (CPU)

1. Clone this repo
2. Run:

cmake . -B build -G Ninja

3. Let CMake generate and run:

cd build && ninja

4. After building you can run (linux & mac):

./face_roi_demo

or (if using windows)

face_roi_demo.exe

Running the examples (Coral TPU)

0. Make sure you got:

• The newest libedgetpu library installed (based on TF 2.16.1); If not sure compile it from source!
• libusb-1.0-0-dev package or installed (only when using Linux)\
• When using Linux your udev configured correctly (is done automatically when compiling from source!)

1. Clone this repo
2. Run:

cmake . -B build -G Ninja -DFACE_DETECTOR_ENABLE_CORAL_TPU=ON

or add this block to your settings.json in vscode:

"cmake.configureSettings": {
    "FACE_DETECTOR_ENABLE_CORAL_TPU": "ON"
}

3. Let CMake generate and run:

cd build && ninja

4. After building you can run (linux & mac):

./face_roi_demo

or (if using windows)

face_roi_demo.exe

Using it in your project as library

Add this to your top-level CMakeLists file:

include(FetchContent)

FetchContent_Declare(
    face_detector
    GIT_REPOSITORY https://github.com/CLFML/Face_Detector.Cpp
    GIT_TAG main
    SOURCE_DIR ${CMAKE_CURRENT_LIST_DIR}/lib/Face_Detector.Cpp
)
FetchContent_MakeAvailable(face_detector)
...
target_link_libraries(YOUR_EXECUTABLE CLFML::face_detector)

Or manually clone this repo and add the library to your project using:

add_subdirectory(Face_Detector.Cpp)
...
target_link_libraries(YOUR_EXECUTABLE CLFML::face_detector)

Building a ROS2 package with Colcon

Before using this library you will need the following packages installed:

• OpenCV
• ROS2
• ROS CV bridge
• Working C++ compiler (GCC, Clang, MSVC (2017 or Higher))
• CMake

Running the examples (Ubuntu, CPU)

1. Clone this repo:

git clone https://github.com/CLFML/Face_Detector.Cpp.git

2. Source your ROS2 installation:

source /opt/ros/jazzy/setup.bash

3. Install the dependencies:

rosdep install --from-paths src -y --ignore-src

4. Build the package:

colcon build --packages-select face_detector

5. Set up the environment:

source install/setup.bash

6. Run the camera node:

ros2 run v4l2_camera v4l2_camera_node

7. In another terminal, run the face detector node (don’t forget to source the setup script):

ros2 run face_detector face_detector_node

8. In yet another terminal, run the viewer:

ros2 run face_detector face_detector_viewer

Aditional documentation

See our wiki...

Todo

• Add language bindings for Python, C# and Java
• Add support for MakeFiles and Bazel
• Add Unit-tests
• Add ROS2 package TPU support
• If needed to crunch ROS2 latency: Do not use nodes, put plugins in a container

License

This work is licensed under the Apache 2.0 license.