Face Alignment

Description

This repository contains a Jupyter notebook which tests and compares 3 simple face alignment Machine Learning algorithms.

The dataset used in these experiments consists of images of families collected from the web. Since most of these images are copyrighted, the dataset is going to remain private.

The face detectors used in these experiments are:

• OpenCV Haar cascade Eyes Detector
• Dlib 5-point Face Landmarking Detector
• YOLOv5Face 5 Facial Keypoints Detector

The detectors are evaluated based on

• Total alignment time
• Total number of faces aligned

Instructions

Setup and execution on Ubuntu 20.04

1. Install necessary tools

sudo apt-get -y install git wget zip bzip2

2. Fetch repository

git clone https://github.com/nbishdev/simple_face_alignment.git

3. Install Miniconda

wget https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
bash Miniconda3-latest-Linux-x86_64.sh
source ~/.bashrc

4. Create a virtual environment for the notebook experiments

conda create -n face

5. Install all the necessary packages

conda install -n face -c conda-forge dlib
conda activate face
pip install -r simple_face_alignment/requirements.txt

6. Execute JupyterLab

jupyter-lab --notebook-dir=simple_face_alignment