Question about ground truth generation

[wl3181]

Hi author,

Thanks for your code and awesome idea.

I have one question: in your code, you generate the gt gaze direction by the segmentation map. Why don't you directly use the ground truth value in OpenEDS2020.

What's the difference and error between OpenEDS2020 gaze gt and results generated by 3D pupil model? I remember in the original 3D pupil model paper, the error is approximately 2 degree