FrontApp is a Node.js web app which uses a face frontalization algorithm to produce frontalized images of faces.
You may also download the code and use it to perform frontalization independently of the web application.
The code has been tested on a machine running Ubuntu 14.04.
This section will explain how to install the web application (on a machine running Linux). Once it's up and running, you will need to compile the frontalization API for it to be available to the web application.
Requirements -
- Node.js (version >= 0.12.7).
- MondoDB (version >= 2.7).
- npm
- ImageMagick
- GraphicsMagick
Once the above requirements are met, go to the web application folder (where package.json is located) and run:
npm install
This will install all of the required Node.js packages listed in package.json.
After that you can run:
node app.js
To start the application.
Once the application is functioning you should compile the frontzlization API.
Requirements -
You will need to install OpenCV yourself. The version of DLib that we use is provided with the code.
To compile the code (assuming that OpenCV is installed system-wide, and that DLib is located at /path/to/app/frontalization_API/dlib-18.15) go to the frontalization_API folder and run:
g++ -fPIC -shared -I./dlib-18.15 `pkg-config opencv --cflags` frontalize_for_lib.cpp -o libfrontalize.so -lpthread -lX11 `pkg-config opencv --libs`
To view the usage and all available options:
$ python Frontalizer.py -h
output:
usage: Frontalizer.py [-h] -ip IMAGES_PATH [-dp DLIB_PATH]
[-cp OPENCV_HAAR_PATH] [-op OUTPUT_PATH] [-write]
[-crop CROP] [-crop_width CROP_WIDTH]
[-crop_height CROP_HEIGHT]
[-crop_x_offset CROP_X_OFFSET]
[-crop_y_offset CROP_Y_OFFSET] [-symthr SYM]
[-blend BLEND] [-debug] [-preproc]
optional arguments:
-h, --help show this help message and exit
-ip IMAGES_PATH The location of the image(s) that will be processed.
You can pass a filename (i.e.: "/home/me/sandra.jpg")
or a glob (i.e.: "/home/me/pics/*.jpg").
-dp DLIB_PATH Path to dlib facial landmarks file, i.e.:
"/home/me/data/shape_predictor_68_face_landmarks.dat".
-cp OPENCV_HAAR_PATH Path to the opencv haarcascade xml files, i.e.:
"/home/me/opencv-3.0.0-rc1/data/haarcascades".
-op OUTPUT_PATH A folder where the results are to be saved.
Pass "-write" if you wish to use this option.
-write CAUTION: If this argument is used, and the -op option
was not passed, existing images will be overwritten
with the frontalized images.
-crop CROP The size (in pixels) of the square area to be processed
when frontalizing the face.c
-crop_width CROP_WIDTH
The width of the cropped face area. If used,
it must be smaller than value used for "crop"
-crop_height CROP_HEIGHT
The height of the cropped face area. If used,
it must be smaller than value used for "crop"
-crop_x_offset CROP_X_OFFSET
The x offset of the cropped face area.
-crop_y_offset CROP_Y_OFFSET
The y offset of the cropped face area.
-symthr SYM Threshold for soft symmetry.
-blend BLEND Blend factor for soft symmetry.
-debug If this argument is used, some debugging information
will be printed.
-preproc If this argument is used, the images will be
preprocessed: the face will be detected and
the image will be cropped to 250x250, with
the face centered in the image.
NOTE: images which are not 250x250 may cause the
frontalization code to crash.
- The original face frontalization algorithm was developed by Dr. Tal Hassner:
Tal Hassner, Shai Harel*, Eran Paz* and Roee Enbar, "Effective Face Frontalization in Unconstrained Images", IEEE Conf. on Computer Vision and Pattern Recognition (CVPR), Boston, June 2015.
* These author names are in alphabetical order due to equal contribution.
- The C++ code used to implement the above algorithm is a modification of the code used in the uniform-lbp project (frontalize.cpp).