Carl M. Gaspar & Oliver G.B. Garrod

import alignfaces as afa

faces_path = "/Users/Me/faces_for_my_study/"
afa.get_landmarks(faces_path)
aligned_path = afa.align_procrustes(faces_path)
afa.get_landmarks(aligned_path)
the_aperture, aperture_path = afa.place_aperture(aligned_path)

To better understand how to write a script for your specific purposes, we direct you to demo 1. Demo 1 also describes how AFA alignment works.

All of these functions depend on reliable detection of facial landmarks, which is provided by the DLIB library. Alignment is based on generalized Procrustes analysis (GPA), which extensively unit tested.

Automatic landmark detection means that it is also easy to separate shape and texture in order to produce various kinds of warped images.

AFA provides functions for two types of face-warping manipulations common in face perception research.

To learn how to do this please see demo 2.

To learn how to do this please see demo 3.

It is highly recommended that you have conda installed, preferably miniconda rather than full fat anaconda.

If you do have conda, then do the following to install:

conda create --name myenv conda-forge::dlib "python>=3.9" scikit-image

conda activate myenv

pip install "alignfaces @ git+https://[email protected]/SourCherries/auto-face-align.git"

This will create a new virtual environment called myenv. You can use another name for that. You'll need to activate this environment using conda activate myenv whenever you want to use AFA. To deactivate, simply type conda deactivate myenv.

Windows users may encounter a problem with plotting. That is a general issue with Matplotlib on Windows. To fix, simply type the following while your myenv is activated:

conda install freetype=2.10.4

In addition to unit-testing critical computations, I evaluated both landmark estimation (DLIB) and the outcome of the entire alignment procedure using various face databases. The results are described here.

If you use this package for your research, please cite the following preprint:

Gaspar, C. M., & Garrod, O. G. B. (2021, November 8). A Python toolbox for Automatic Face Alignment (AFA). Retrieved from psyarxiv.com/erc8a

DOI:

10.31234/osf.io/erc8a

This module is under an Apache-2.0 license.