Feature: try out different models and training strategies using notebooks and document them

[KarinePistili]

We would like to test out different AI models and methods for the eye tracking process. We ask you to save your test and findings on notebooks inside the repository. The idea is to make simple proof of concepts so we can comparte the different methods.

[gitone912]

can i do this one , i am good at modelling

[KarinePistili]

Yes you can. I'll assign it you @gitone912 . If you have any questions feel free to send them here ;)

[gitone912]

sure thanks a lot @KarinePistili

[gitone912]

hey @KarinePistili i tried different models on the data we generate through calibration process , the accuracy and error rates are all depends on how perfect the user calibrate so their data dispersion becomes less.
the question arises here in my mind is what exact goal are we trying to accomplish using different models because more or less accuracy and errors are dependent on how user interact in test so even if we change various model , superiority of data will always be high
please clear my doubt
thankyou

[Dhruvanshu-Joshi]

Hey @KarinePistili , I’d love to take this up if the current work is halted.

I propose to test different AI models for eye tracking and documenting the results in Jupyter Notebooks. The idea is to see which approach works best for accuracy, speed, and reliability. I’ll start with:

• DeepGaze (CNN-based): A solid deep-learning baseline that predicts gaze direction from images.

• GazeNet (Deep-learning): More advanced, designed specifically for gaze tracking, so it might improve accuracy.

• OpenFace 2.0 (Feature-based): Lightweight and interpretable, which could make it more efficient.

I’ll begin with DeepGaze and iterate from there. This should help us find a model that improves tracking while staying compatible with the current setup.

Please let me know if I am in the current direction so that I can open a PR. Looking forward to hearing from you.

Thank you