👁️ Peekr: Browser-based Eye Tracking

Peekr is a lightweight, browser-compatible webcam eye tracking module. It runs entirely in the browser — no installation, no data sent to a server.

🎓 Background

Peekr was developed by Aryaman Taore, a visual neuroscientist and machine learning engineer. The project emerged during Aryaman's PhD at the University of Auckland and continued through his postdoctoral research at Stanford University. Peekr's underlying ML model was trained on 268,000+ image frames from 264 participants recruited via Prolific. Each participant used their own personal setup.

Accuracy (after calibration):

• Mean horizontal error: 1.53 cm (~1.75° visual angle)
• Mean vertical error: 2.20 cm (~2.52° visual angle) These results were obtained from 30 randomly selected participants using their own setups, with no supervision. Each participant followed a stimulus on screen after completing a simple 5-dot calibration. The calibration consisted of four dots in the corners and one in the center of the screen. After this, a linear fit was applied separately to the x and y axes to adjust the gaze predictions.

🧪 Demo (with Calibration Panel)

I have attached a public/index.html demo file. To run this demo, follow these steps:

1. First, build the project by running the following command:

   npm install
   npm run build

2. Then, run the demo using npx with the following command:

   npm run serve

🧠 Available Functions

Peekr.initEyeTracking({ onReady, onGaze })

Initializes the webcam, loads the ONNX model, and sets up gaze detection.

• onReady: Called once the model is loaded and initialized.

• onGaze(gaze): Called every frame with the gaze prediction:

  gaze.output.cpuData = [x, y]; // Both values in range ~[0, 1]

Peekr.runEyeTracking()

Starts real-time gaze prediction.

Peekr.stopEyeTracking()

Stops webcam and gaze processing.

📁 Files Included

| File | Purpose | | ------------- | ------------------------------- | | index.js | Main entry module | | worker.js | Off-main-thread gaze processing | | peekr.onnx | Gaze detection model | | .wasm files | ONNX Runtime Web WASM backend |

🚀 Load via npm

npm install peekr

Then:

import * as Peekr from 'peekr';

📦 Publishing

To build and publish:

npm run build
npm publish --access public

Make sure your dist/ includes the model, worker, and all necessary WASM files.

🧠 Credits

Built by Aryaman Taore at Dakin Lab and Stanford Brain Development & Education Lab.

🗪 License

MIT