GazeCapture RunwayML

This repository brings the pre-trained model from Eye Tracking for Everyone into python and RunwayML.

It allows for a server to be spun up in a docker container that performs real- time gaze estimation from a video stream. It works with any webcam.

I have used this model successfully in my project Presence - a kinetic sculpture that reacts to a users gaze.

The server:

Install Docker.

Build the docker container in a tag:

docker build . -t gaze

Launch the container:

./start_container.sh

The client:

The client opens a webcam video feed and sends it in a stream to the server, getting gaze positions back.

Install dependencies:

pip install opencv-python zmq
    

Launch the client:

python test_client.py