Skip to content

Eye Tracking Algorithm Optimization Based on Low‐Resolution Cameras ‐ Sitam Meur

Jump to bottom
KarinePistili edited this page Jan 16, 2026 · 2 revisions

GSoC'24 — Eye Tracking Algorithm Optimization Based on Low-Resolution Cameras

This project aims to improve eye-tracking algorithms by exploring new techniques, optimizing existing parameters, and integrating JavaScript enhancements for seamless front-end integration. The main goal is to achieve real-time eye-tracking capabilities. Once the optimization process is complete, the algorithms will be carefully integrated into the project for usability testing, ensuring compatibility and smooth operation within the existing system's framework.

🌐 Official GSoC Project Page

🔗 Google Summer of Code 2024 — Web Eye Tracker Project

👨‍💻 Contributor

  • Sitam Meur
    • Role: GSoC Contributor – Student Developer
      Education: Bachelor of Technology in Computer Science, Meghnad Saha Institute of Technology, Kolkata
    • GitHub Profile
    • LinkedIn Profile

🧑‍🏫 Mentors

🧩 Project Overview

The Web Eye Tracker project aims to create an accessible, open-source, and browser-compatible eye tracking tool capable of operating on low-resolution cameras.
The system enables gaze tracking, calibration, model selection, and real-time visualization, enhancing usability testing and research capabilities across diverse devices.

The project was developed in two major components:

1. 🖥️ Backend — eye-tracker-api

Repository: ruxailab/eye-tracker-api
Stack: Flask, Python, Docker, Streamlit, Google Cloud Run

Core Responsibilities:

  • Implemented multiple regression models for gaze prediction:
    • Linear, Ridge, Lasso, Elastic Net, Bayesian Ridge, and SGD Regressor.
  • Integrated model selection and calibration endpoints for flexible experimentation.
  • Built data preprocessing, model persistence, and evaluation metric pipelines (MAE, RMSE, R²).
  • Developed a Streamlit-based visualization tool for analyzing raw calibration data, metrics, and regression outputs.
  • Implemented logging, testing, and configuration management.
  • Containerized the service using Docker for easy deployment on Google Cloud Run.

Links:

2. 💡 Frontend — web-eye-tracker-front

Repository: ruxailab/web-eye-tracker-front
Stack: Vue.js, Vuetify, TensorFlowJS, JavaScript, Firebase

Key Deliverables:

  • Implemented the front-end dashboard to:
    • Manage calibration and gaze prediction sessions.
    • Enable model selection and real-time gaze visualization.
    • Display regression outputs and metrics dynamically.
  • Integrated TensorFlowJS for in-browser ML inference.
  • Connected front-end with backend via REST API endpoints.
  • Improved UI/UX for accessibility and real-time feedback.
  • Conducted internal testing and usability validation using low-resolution webcams.

Links:

🗂️ Project Management

Resource Link
📊 Project Proposal RUXAILAB GSoC Project Proposal
📈 Progress Sheet GSoC Progress & Work Log
🔄 Backend PR PR #26
🧠 Frontend Commits web-eye-tracker-front commits

📚 Documentation & Research Notes

Throughout the project, several technical notes and internal guides were created:

  • 🧩 Model Comparison Report: Regression, Ensemble, and Neural architectures
  • ⚙️ Hyperparameter Optimization Notes: Grid Search, Cross-validation, and Bayesian tuning
  • 🎛️ Calibration Study: Analysis of manual calibration limitations and ML-based improvements
  • 🧪 Testing Notes: Model performance analysis, overfitting mitigation, and validation logs

🏁 Outcome

The project successfully delivered a fully integrated web eye-tracking system, with modular backend APIs and interactive front-end visualizations.

Key Achievements:

  • ✅ Complete end-to-end integration between backend and frontend.
  • ✅ Flexible model selection & evaluation system.
  • ✅ Streamlit visualization for calibration and results analysis.
  • ✅ Performance testing with real camera input.

🧠 Technical Highlights

  • Addressed overfitting in the Y-coordinate regression model via regularization and validation techniques.
  • Designed a modular model interface allowing seamless switching between multiple regressors.
  • Implemented TensorFlowJS model integration for live browser inference.
  • Conducted comparative analysis between traditional regressors and advanced models (MLP, CNN, Mediapipe).
  • Proposed future enhancements for deep learning–based calibration and head-pose correction.

🏆 GSoC Summary

Item Description
Organization RUXAILAB
Program Google Summer of Code 2024
Contributor Sitam Meur
Project Eye Tracking Algorithm Optimization Based on Low-Resolution Cameras
Technologies Python, Javascript, Flask, Vue.js, Scikit-learn, Firebase, Streamlit, Tensorflow.js
Topics Front-end, Javascript, Eye Tracking, Algorithm Optimization, Artificial Intelligence (AI)
Duration May – August 2024

✨ Acknowledgements

Special thanks to mentors Vinicius, Marc, and Karine Pistili for their invaluable support and continuous guidance throughout the GSoC journey.

📚 Useful Links

Resource Link
🧩 Backend Repository eye-tracker-api
💻 Frontend Repository web-eye-tracker-front
📄 Final PR #26
📊 Project Proposal RUXAILAB GSoC Project Proposal
📈 Progress Sheet GSoC Progress & Work Log

Submitted as part of Google Summer of Code 2024 – Final Work Proof
© 2025 RUXAILAB • Developed by Sitam Meur

🏠 RUXAILAB

☀️ GSoC

Clone this wiki locally