Smooth Pursuit

This page makes publicly available the source code and the data for detection and analysis of smooth pursuit eye movements. For our hand-labelling tool, please see this link.

For questions, feedback, etc., please contact us via mail at < firstname.lastname > @tum.de : Mikhail Startsev, Ioannis Agtzidis, Michael Dorr.

Results

Source code and data

• deep_eye_movement_classification_package.zip contains the source code and data of our 1D CNN-BLSTM approach. See README for more details. If you use this algorithm or its derivatives, please cite our paper:

  @article{startsev2018cnn,
  author="Startsev, Mikhail and Agtzidis, Ioannis and Dorr, Michael",
  title="1D CNN with BLSTM for automated classification of fixations, saccades, and smooth pursuits",
  journal="Behavior Research Methods",
  year="2018",
  month="Nov",
  day="08",
  issn="1554-3528",
  doi="10.3758/s13428-018-1144-2",
  url="https://doi.org/10.3758/s13428-018-1144-2"
  }

  A read-only free access version of the paper can be found here.
  
  The eye movement labels produced by all the algorithms that were evaluated for the publication above can be downloaded here. The archive includes the results for our latest configuration of the 1D CNN-BLSTM architecture (see the ECVP'18 reference below).
  
  Part of this work was also presented as a demo at ETRA'18 (see the abstract, video explanation, and poster). You can cite it as follows:

  @inproceedings{startsev2018deep,
  author = {Startsev, Mikhail and Agtzidis, Ioannis and Dorr, Michael},
  title = {Deep Learning vs. Manual Annotation of Eye Movements},
  booktitle = {Proceedings of the 2018 ACM Symposium on Eye Tracking Research \& Applications},
  series = {ETRA '18},
  year = {2018},
  isbn = {978-1-4503-5706-7},
  location = {Warsaw, Poland},
  pages = {101:1--101:3},
  articleno = {101},
  numpages = {3},
  url = {http://doi.acm.org/10.1145/3204493.3208346},
  doi = {10.1145/3204493.3208346},
  acmid = {3208346},
  publisher = {ACM},
  address = {New York, NY, USA},
  keywords = {event detection, eye movement classification, smooth pursuit},
  }

  
  For ECVP'18, we evaluated different architecture configurations, further improving the performance of our model. The outputs of this model are already added to the deep_eye_movement_classification_package.zip archive. You can find the ECVP'18 presentation here. The citation for this work is located below. Please do not reference this work in isolation, but together with the "1D CNN with BLSTM for automated classification of fixations, saccades, and smooth pursuits" paper from above.

  @inproceedings{startsev2018sequence,
  title = {Sequence-to-sequence deep learning for eye movement classification},
  author = {Startsev, Mikhail and Agtzidis, Ioannis and Dorr, Michael},
  booktitle = {Proceedings of the European Conference on Visual Perception},
  year = {2018},
  note = {(In press)}
  }

• sp_tool.zip contains source code for multi-observer based smooth pursuit detector. See README for more details. If you use this algorithm or its source code in your work, please cite

  @inproceedings{agtzidis2016smooth,
  title={Smooth pursuit detection based on multiple observers},
  author={Agtzidis, Ioannis and Startsev, Mikhail and Dorr, Michael},
  booktitle={Proceedings of the Ninth Biennial ACM Symposium on Eye Tracking Research \& Applications},
  pages={303--306},
  year={2016},
  organization={ACM}
  }

• larsson_reimplementation.zip additionally provides a reimplementation of the baseline SP detection algorithm described in Larsson et al., Biomedical Signal Processing and Control, 2015.


• GazeCom.zip contains raw gaze data, hand-labelled ground truth, and outputs for three SP labelling algorithms in the ARFF format (see here for more details) for the GazeCom data set (check the Stimuli section for the respective video files).
  
  
  If you use the data published here, please cite the website as

  @misc{sp-detection-site,
  author = {Startsev, Mikhail and Agtzidis, Ioannis and Dorr, Michael},
  title = {Smooth Pursuit},
  howpublished = {\url{http://michaeldorr.de/smoothpursuit/}},
  year = {2016}
  }.
  

  Data were published in connection with an ECEM 2017 poster:

  @inproceedings{StAgDo17,
  title = {Manual \& Automatic Detection of Smooth Pursuit in Dynamic Natural Scenes},
  author = {Startsev, Mikhail and Agtzidis, Ioannis and Dorr, Michael},
  booktitle = {Proceedings of the European Conference of Eye Movements},
  year = {2017},
  }

Label smoothing (as post-processing)

@inproceedings{StLeDo17,
title = {Optimizing clustering-based smooth pursuit detection},
author = {Startsev, Mikhail and Lee, Albert Tae-Young and Dorr, Michael},
booktitle = {Proceedings of the European Conference on Visual Perception},
year = {2017},
note = {(In press)}
}

Evaluation

Submission

Submission format