Fovea SDK

Foveated Streaming Engine for High-Performance Screen Sharing.

Overview

Fovea uses WebAssembly (Rust) to perform "Foveated Rendering" on video streams entirely on the client side. It detects changes in the screen and only sends high-resolution updates for the active region (ROI), while blurring the static background. This results in massive bandwidth savings (up to 90%) for screen sharing applications.

Installation

npm install fovea-sdk

(Note: During private beta, install from local path or private registry)

Usage

1. Electron (Node.js)

const FoveaEngine = require('fovea-sdk');
const path = require('path');

async function start() {
    const engine = new FoveaEngine({ blurAmount: '12px' });
    
    // Initialize with local WASM file
    await engine.initialize({ 
        wasmPath: path.join(__dirname, 'node_modules/fovea-sdk/fovea_rs/pkg/fovea_wasm_bg.wasm') 
    });

    function loop() {
        // Process video frame -> canvas
        const result = engine.process(videoElement, outputCanvas);
        
        if (result && result.type === 'foveated') {
             console.log(`ROI: ${JSON.stringify(result.bbox)}`);
        }
        requestAnimationFrame(loop);
    }
}

2. Browser (React/Vue/Vanilla)

Copy the .wasm file to your public assets folder first.

import FoveaEngine from 'fovea-sdk';

const engine = new FoveaEngine();

// Initialize by fetching WASM from public URL
await engine.initialize({ 
    wasmPath: '/assets/fovea_wasm_bg.wasm' 
});

// ... use engine.process() in your render loop

API

new FoveaEngine(config)

• config.blurAmount: CSS blur string (default '8px')
• config.diffThreshold: Sensitivity (0-255) for change detection (default 30)

async initialize(options)

• options.wasmPath: Path (Node) or URL (Browser) to the .wasm binary.
• options.wasmBuffer: ArrayBuffer of the WASM binary (optional alternative).

process(video, canvas)

• video: Source HTMLVideoElement
• canvas: Target HTMLCanvasElement (where the foveated effect is drawn)
• Returns: { type, bbox, isKeyframe }

🤝 Contributing

Want to improve the engine? We welcome optimizations! Since the core is written in Rust, please see our Contribution Guide for build instructions.

License

MIT