JsPsych integration - Problem reaching high accuracy

[Naveh-Ayal]

Dear WebGazer team,
I am using your great eye-tracking for a basic image similarity experiment as part of a larger research project.
I used the JsPsych plugins, and took most of the code from the WebGazer examples inside the WebGazer-JsPsych Github.
The accuracy is reletavely low (even in the example.html compared to the example presented in the brown.edu page.

I deployed the experiment (and the WebGazer-JsPsych example) over http and over https, and still I do not get the accuracy I was hoping for.
I was wandering if you know of a decrease in accuracy when using throgh JsPsych and if there is anything you believe I can do to improve the accuracy.

Thank you in advance,
Naveh

[jeffhuang]

Hi @Naveh-Ayal, the first two things I would try are 1) test out the latest WebGazer version by doing cloning the version on this repository, and 2) making sure the user's head is fairly stable during calibration and does not move substantially during the trial.

[Naveh-Ayal]

Dear @jeffhuang,

Thank you very much for your quick and helpful reply.

For some context:
I would like to use WebGazer for a relatively simple fixation check before each similarity rating trial. Specifically, participants will be asked to maintain fixation on a central cross, and the trial will begin only if stable fixation (within a preset radius) is maintained for 500 ms, within a 10-second window.

To the best of your knowledge, would you consider WebGazer a suitable solution for this kind of task?

If so, I would greatly appreciate your advice on a few follow-up questions:

1. Based on your recent testing, what level of tracking error (e.g., degrees of visual angle) should I expect when using the latest version of WebGazer under good conditions?

2. Is there an updated integration of the newest WebGazer version into jsPsych that you would recommend using?

3. If not, do you have any suggestions for the best way to approach the integration?
   In the current jsPsych setup, there are dedicated plugins for camera initiation, calibration, and validation, alongside the WebGazer extension and the webgazer.js script.
   Would you recommend maintaining a similar architecture for integrating the latest version, or are there newer best practices based on WebGazer’s recent improvements?

Thank you again for your time and support!
Naveh

[jeffhuang]

WebGazer is frequently being updated so we can't run formal tracking error tests, but around half an inch of error is fairly common. We're upstream from jsPsych, so questions about jsPsych may be best directed to that team. You could try out the current WebGazer in the repository, and see if it performs as you expect standalone, and if you feel like that's better than what's in jsPsych, then I'd suggest trying to adopt it in jsPysch.