Face Landmarks Detection In Your Android App — Part 1

• Download the source code (I’m using v19.4).
• Install homebrew to manage the development tools or packages.
• Install the C/C++ compiler.
  e.g. brew install llvm
• Install cmake. It uses cmake as the build system to generate the Makefile.
  e.g. brew install cmake
• Install xquartz. It uses x11/xlib as GUI framework.
  e.g. brew cask install xquartz
• Install lapack. It uses Linear Algebra PACKage to compute the matrix operation.
  e.g. brew install lapack
• Download the pre-trained model.
  e.g. http://dlib.net/files/shape_predictor_68_face_landmarks.dat
• Prepare the photos with clear faces inside.

The sample code I’m interested is face_landmark_detection_ex.cpp which is located in the examples directory. And it would be compiled to an executable binary file that could take the trained model and photos as inputs.

Go into the examples folder and type:

cd examples && mkdir build && cd build && cmake .. && cmake — build .

If it is successfully compiled, you should see many generated executable binary files in the examples/build/ directory.

The sample relies on the xquartz server to display the result. So remember to launch XQuartz.app first (located at /Applications/Utilities/XQuartz.app) and keep it opened. Then run the sample to see how it performs:

# 1st parameter is the trained model; 2nd parameter is the photo
./face_landmark_detection_ex \
    shape_predictor_68_face_landmarks.dat \
    ../faces/*.jpg

If you encounter DLIB_NO_GUI_SUPPORT compiling error like me. It is probably because the generated dlib/build/config.h header file is not used. To solve it you simply copy the dlib/build/config.h to dlib/config.h.

The cmake I’m using doesn’t support interactive debugging. To do further debugging, I add debugging message to know what flags are turned on or off in the CMakeLists.txt file.

# In CMakeLists.txtmessage(STATUS "The log message.")

Run cmake and dump the debugging message.

cmake . --debug-output -Wdev

The result looks pretty cool to me. In the next article, I will be trying to cross-compile the library to be used by an Android app.

Original paper: pdf

Watch on YouTube