gazeclassify 0.9.1

Algorithmic eye-tracking analysis

GazeClassify

PiPy package to algorithmically annotate eye-tracking data.

Special thanks to Pixellib for providing algorithms and to PySport for inspiration with the domain model.

---

What is GazeClassify?

GazeClassify provides automatized and standardized eye-tracking annotation. Anyone can analyze gaze data online with less than 10 lines of code.

Exported csv will contain distance from gaze (red circle) to human joints (left image) and human shapes (right image) for each frame.

frame number	classifier name	gaze_distance [pixel]	person_id	joint name
0	Human_Joints	79	0	Neck
...	...	...	...	...
0	Human_Shape	0	None	None
...	...	...	...	...

Run on example data

from gazeclassify import Analysis, PupilLoader, SemanticSegmentation, InstanceSegmentation
from gazeclassify import example_trial

analysis = Analysis()

PupilLoader(analysis).from_recordings_folder(example_trial())

SemanticSegmentation(analysis).classify("Human_Shape")
InstanceSegmentation(analysis).classify("Human_Joints")

analysis.save_to_csv()

Run on your own data

Capture eye tracking data from a Pupil eye tracker. Then, export the data using Pupil software. You will get a folder with the exported world video and the gaze timestamps. Finally, let gazeclassify analyze the exported data:

from gazeclassify import Analysis, PupilLoader, SemanticSegmentation, InstanceSegmentation

analysis = Analysis()

PupilLoader(analysis).from_recordings_folder("path/to/your/folder_with_exported_data/")

SemanticSegmentation(analysis).classify("Human_Shape")
InstanceSegmentation(analysis).classify("Human_Joints")

analysis.save_to_csv()

Citation

Please cite this paper in your publications if GazeClassify helps your research.

@inproceedings{10.1145/3450341.3458886,
  author = {M\"{u}ller, Daniel and Mann, David},
  title = {Algorithmic Gaze Classification for Mobile Eye-Tracking},
  year = {2021},
  booktitle = {ACM Symposium on Eye Tracking Research and Applications}
}