OpenGaze: Web Service for OpenFace Facial Behaviour Analysis Toolkit

Overview

OpenFace is a fantastic tool intended for computer vision and machine learning researchers, the affective computing community, and people interested in building interactive applications based on facial behavior analysis. OpenFace is the first toolkit capable of facial landmark detection, head pose estimation, facial action unit recognition, and eye-gaze estimation with available source code for running and training the models.

Even though the documentation of this tool is very detailed and organized, it is still problematic to use in web services because of its C++ codebase and CLI interface.

OpenGaze is a single endpoint RESTful web API service with HTTP Basic Authentication developed with FastAPI framework that uses the FaceLandmarkImg executable of OpenFace and provides a web API that responds with crucial eye-gaze and head-pose related fields.

Installation

OpenFace Installation

Follow the installation instruction of OpenFace to install it first with all the dependencies. Platform based installation instructions are as follows-

The install script for Unix-like systems only work for Debian-based systems including ubuntu. For other distros, please install the dependencies manually and then compile the code following the Advanced Ubuntu Installation section.

At the end of third step of Actual OpenFace installation following command can be used to install the executables in the system.

make install

If make install is not used, FACE_LANDMARK_IMG_EXEC_COMMAND environment variable in .env file will need to be updated with the absolute path of the FaceLandmarkImg executable. e.g., /home/user/OpenFace/build/bin/FaceLandmarkImg

Run OpenGaze

Clone this repo

git clone https://github.com/nsssayom/OpenGaze.git
cd OpenGaze

Install virtualenv package for your OS. For Debian-based systems-

sudo apt update
sudo apt install python3-virtualenv

Create and activate a virtual environment.

virtualenv env
source env/bin/activate

Install dependencies using pip.

pip install -r requirements.txt

Run OpenGaze

uvicorn main:app --reload 

Usages

Now, test the API with any API testing tool. cURL snippet is following

curl --request POST \
  --url http://localhost:8000/ \
  --header 'Authorization: Basic <RandomHeader>' \
  --header 'Content-Type: application/json' \
  --data '{
	"image_url": "https://wamu.org/wp-content/uploads/2019/05/Sun-article.jpeg"
}'

use API_KEY value in .env file as username and API Secret as password for basic authentication.