face-collector

Collect faces from images and/or videos.

Samples

Collect faces from images

$ ./face-collector <images_dir>

6 photos in <images_dir>

faces collected from above photos

Sample photos taken from Pictures Of People @ PEXELS

Collect faces from video

$ ./face-collector capoiera_street_dance.mp4

List of the first 20 images of collected faces.

Collect and rotate the face

Run with --rotate option, you can rotate face so that the two eys are horizontal.

$ ./face-collector capoiera_street_dance.mp4 --rotate

List of the first 20 images of collected faces

Collect and expand face detection rectangle

Run with --expansion option, you can expand face detection rectangle by dlib based on interpupillary distance.

$ ./face-collector capoiera_street_dance.mp4 --expansion 2.0

List of the first 20 images of collected faces

Collect faces from images and/or videos

Save images and/or videos in a directory. And run face-collector with the direcotry name.

$ ./face-collector <target_dir>

Sample video (capoiera_street_dance.mp4) taken from Capoiera Street Dance @ PEXELS VIDEOS

Collect faces from webcam

Run witn --webcam <device-id> option, you can collect faces from built-in camera or webcam.

$ ./face-collector --webcam 0

Install requirements

• Python 2.7.x
• Dlib
• ffprobe including in ffmpeg (version >= 2.1)

Installation

$ git clone https://github.com/knjcode/face-collector.git
$ cd face-collector
$ pip install -r requirements.txt
$ wget http://dlib.net/files/shape_predictor_68_face_landmarks.dat.bz2
$ bzip2 -d shape_predictor_68_face_landmarks.dat.bz2

If you are using pyenv and you can not install dlib, try reinstalling python as in the following steps.

$ pyenv uninstall 2.7.13
$ env PYTHON_CONFIGURE_OPTS="--enable-shared" pyenv install 2.7.13

Usage

Run face-collector with target (image or video or directory including images and/or videos).

$ ./face-collector.py <target_path>

Collected faces are saved in faces directory by default.

Options

• --output (default faces)

  Set output directory

  $ ./face-collector.py <target_path> --output /path/to/output
  

• --confidence (default 0.6)

  Set a threshold for face detection confidence

  $ ./face-collector.py <target_path> --confidence 0.9
  

• --resize

  Specify the pixel dimensions of the face images.

  $ ./face-collecotr.py <target_path> --resize 64
  

• --prefix (default face)

  Specify the filename prefix

  $ ./face-collector.py foo.jpg --prefix sample
  image: foo.jpg
  Saved: faces/sample00000.png
  

• --zerofill (default 5)

  Specify the number of digits for zero-padding

  $ ./face-collector.py foo.jpg --zefofill 3
  image: foo.jpg
  Saved: faces/face000.png
  

• --rotate (default False)

  Rotate face so that the two eys are horizontal

• --expansion (default 1.5)

  Expand face detection frame by dlib based on interpupillary distance.

  $ ./face-collector.py <target_dir> --expansion 2.0
  

  expansion value must be at least 1.0

• --frame-skip (default 30)

  Specify frame skipping of video

  $ ./face-collector.py video.mp --frame-skip 30
  

  When 0 or 1 is specified, face detection is performed for all frames.

• --webcam <device-id>

  Collect faces from built-in camera or webcam

  $ ./face-collector.py --webcam 0
  

• --imgcat

  Display result image for iTerm2 (like a imgcat)

License

MIT