Face-Averager

Collects, aligns and average faces from a directory of images or a video. Examples below.

Dependencies

• Python >= 3.7
• cmake (apt install cmake)
• dlib (pip3 install dlib)
• scikit-image (pip3 install scikit-image)

How it works

I've done a writeup about it on my blog here.

Workflow

The included manager.sh bash file will ensure you have all necessary dependacies and scripts. The manager.sh will then call facePlucker.py to get the best unique faces from a video file. Then, or otherwise faceShaper.py gets called to calculate with a NN model, 68 different places on the face to help align and average. This step is necessary to prevent vomit output. Stacking hundreds of images with relative transparency would quickly get noisy and unless each photo's face are perfectly aligned and centered this is necessary. Finally faceAverage.py (installed by manager.sh) will grab the faces and the spatial information to average and align the Delaunay triangles. The face is then written to output.png.

Tip

You don't necessarily have to hit the streets to get hundreds of images, go to Google Images and load a whole page of results (for example: nyc male or australian bogan). Right click > Save page as. Give the photo folder of that download to the script and you'll get your average with little effort.

Example output / usage

Directory of images

$ ./manager.sh australianMales/

[+] Model already downloaded
[+] Already have faceAverage.py
[+] And with that, the environment is setup

[+] Averaging faces from pictures in: australianMales/

[+] Face triangulation is complete!                       
average.png written, enjoy!

Video file

$ ./manager.sh sydneyWalk.mp4

[+] Model already downloaded
[+] Already have faceAverage.py
[+] And with that, the environment is setup

[+] Averaging faces from video: sydneyWalk.mp4

[-] Extracting every 5th frame, may take a bit, make yourself a cup of tea
[+] Face triangulation is complete!                            
average.png written, enjoy!