TensorFlow.js Monorepo (Web + React Native)

This monorepo contains two minimal apps that demonstrate TensorFlow.js:

• web/ — Vite + React (runs @tensorflow-models/mobilenet in browser)
• mobile/ — Expo React Native (runs @tensorflow/tfjs-react-native + mobilenet on-device)

The goal is to show breadth: same ML concept (image classification) implemented in both web and mobile with shared UI ideas.

Folder Structure

tfjs-monorepo/
├─ web/           # Vite + React web app
└─ mobile/        # Expo React Native app

---

Web App (web/)

Features

• Upload an image, run mobilenet inference in browser (WebGL/CPU)
• Display top-3 predictions

Run

cd web
npm i
npm run dev

Then open the printed localhost URL.

---

Mobile App (mobile/)

Features

• Pick image from camera roll
• Load mobilenet via @tensorflow/tfjs + @tensorflow/tfjs-react-native
• Run on-device inference and show top-3 predictions

Run (Expo)

cd mobile
npm i
# Prebuild native code (ensures GL + deps are installed)
npx expo prebuild
# Start the app
npx expo start

Open on iOS/Android via the Expo app or simulator.

Notes:

• tfjs-react-native uses a custom backend; first load may take a few seconds.
• Ensure you accept camera roll permissions when picking an image.

---

Next Steps

• Swap mobilenet with a custom TFJS graph model
• Add live camera preview classification
• Share preprocessing utils between web and mobile (e.g., as a small local package)
• Add e2e tests (Cypress for web, Detox for mobile)

MIT © 2025