Skip to content
Switch branches/tags


A deep learning framework based on Tensorflow for the training of high performance gaze estimation.

Please note that though this framework may work on various platforms, it has only been tested on an Ubuntu 16.04 system.

All implementations are re-implementations of published algorithms and thus provided models should not be considered as reference.

This framework currently integrates the following models:


Eye region Landmarks based Gaze Estimation.

Seonwook Park, Xucong Zhang, Andreas Bulling, and Otmar Hilliges. "Learning to find eye region landmarks for remote gaze estimation in unconstrained settings." In Proceedings of the 2018 ACM Symposium on Eye Tracking Research & Applications, p. 21. ACM, 2018.


Deep Pictorial Gaze Estimation

Seonwook Park, Adrian Spurr, and Otmar Hilliges. "Deep Pictorial Gaze Estimation". In European Conference on Computer Vision. 2018

To download the MPIIGaze training data, please run bash get_mpiigaze_hdf.bash

Note: This reimplementation differs from the original proposed implementation and reaches 4.63 degrees in the within-MPIIGaze setting. The changes were made to attain comparable performance and results in a leaner model.

Installing dependencies

Run (with sudo appended if necessary),

python3 install

Note that this can be done within a virtual environment. In this case, the sequence of commands would be similar to:

    mkvirtualenv -p $(which python3) myenv
    python3 install

when using virtualenvwrapper.


Tensorflow is assumed to be installed separately, to allow for usage of custom wheel files if necessary.

Please follow the official installation guide for Tensorflow here.

Getting pre-trained weights

To acquire the pre-trained weights provided with this repository, please run:

    bash get_trained_weights.bash

Running the demo

To run the webcam demo, perform the following:

    cd src

To see available options, please run python3 --help instead.


  • datasets/ - all data sources required for training/validation/testing.
  • outputs/ - any output for a model will be placed here, including logs, summaries, and checkpoints.
  • src/ - all source code.
    • core/ - base classes
    • datasources/ - routines for reading and preprocessing entries for training and testing
    • models/ - neural network definitions
    • util/ - utility methods


Gaze Estimation using Deep Learning, a Tensorflow-based framework.




No releases published


No packages published