brownhci/WebGazer

Webgazer Prediction reverts back to center on extremes.

Open

#263 opened on Jun 28, 2022

View on GitHub
 (1 comment) (0 reactions) (0 assignees)HTML (3,412 stars) (520 forks)batch import
help wanted

Description

The issue is when I move my eyes to the extremes of my vision after a certain point the webgazer prediction reverts back to center. I have depicted the problem via this graph as well.

  • In this graph I start looking from the center and gradually move my eyes to the left and as you can see after a certain point the predictions starts to center again.
  • I checked this is not only the problem in my build but same can be observed in the official demo on the webgazer website.
  • I also checked that it is not a problem with mediapipe facemesh since it is successfully able to get the eye crops and also detect the landmarks of correctly.

I am wondering if there is a certain threshold in prediction which leads to this. Anyways Any ideas to why this might be the case ?

Contributor guide

Tech stack
javascripthtml
Domain
frontend
Issue type
bug
DifficultyEstimated implementation difficulty for a new contributor, from 1 for very small changes to 5 for expert-level work.
3
Estimated timeA rough time range for an experienced contributor to investigate, implement, test, and prepare a pull request.
1-2 days
Activity statusHow available the issue appears right now: fresh, active, stale, blocked, or waiting on maintainer input.
stale
ClarityHow clearly the issue explains the expected change, acceptance criteria, and next step.
mostly clear
Prerequisites
JavaScriptWebgazer library internals
Newbie friendlinessA 1-100 score estimating how approachable this issue is for first-time contributors.
30
Research direction
The issue appears to be related to the prediction function in Webgazer. Investigate the ridge regression implementation in `src/regression.js` or similar. Look for any clamping or normalization that might force predictions back to the center when eye positions are extreme. Compare the prediction logic with the calibration data to identify the threshold causing this behavior.