The primary function of this code is to provide tools to estimate gaze direction given Aria eye tracking camera images. The estimates are generated from a pytorch model trained on Aria device data.
Additonally, this code contains tools to visualize the raw and derived gaze data.

Installation & Usage

Running from the GitHub repository

# Install requirements
python3 -m pip install easydict pyyaml torch torchvision projectaria_tools

Run the demo

cd projectaria_eyetracking
python3 model_inference_demo.py --vrs sample.vrs

Enjoy!
A file general_eye_gaze.csv contains the EyeGaze estimates that are compatible with projectaria_tools API.

Install and run on your system

python3 -m ensurepip
python3 -m pip install .
eyetracking_demo --vrs sample.vrs

License

The model is licensed under the Apache 2.0 license.

Contributing

See contributing and the code of conduct.