Face-alignment-in-3000fps

This project is a C++ reimplementation of face alignment in 3000fps in the CVPR 2014 paper: Face Alignment at 3000 FPS via Regressing Local Binary Features. .

Update openMP support !!!

I modify my code to support openMP. You can use it in GCC(Linux) or in VS (Windows).

If you use it in Linux, you should comment or uncomment FIND_PACKAGE( OpenMP REQUIRED) in CmakeLists.txt.

If you use it in Windows, you can directly use it.

VS project

I add a VS project.

Usage

1. Download datasets and get Path_Images.txt as jwyang/face-alignment.

2. To compiler the program: go to folder build and

   cmake .

   make

3. To train a new model: set global parameters, model path, train database name in LBF.cpp. Use "LBF.out TrainModel".

4. To test a model on dataset: set model path, test dataset name in LBF.cpp. Use "LBF.out TestModel".

###Model I have trained a model on AFW, HELEN,LFPW dataset. You can download it from here or google drive.

FAQ

• How to get the bounding box of image ?

  I use the face detector in OpenCV to get the bounding box.You can use any detector to get the bounding box but you must provide a bounding box of similar measure with the training data.

• How about the liblinear?

  I add the liblinear source code as the project code. So you can directly compiler this project and don't need to consider to compiler this library.

Contact

If you have any question, you can create an issue on GitHub. Or you can email yulequan@zju.edu.cn

Reference Project

• soundsilence/FaceAlignment

• jwyang/face-alignment