Archived, no longer maintained
Projects

MyPokemonScanner

A privacy-first, offline-capable Pokemon card scanner running 100% client-side in the browser with YOLO11n and MobileCLIP-S2.

MyPokemonScanner browser card scanner demo
Archived, no longer maintained
  • YOLO11
  • MobileCLIP
  • ONNX Runtime Web
  • WebGPU
  • React
  • Vite

MyPokemonScanner is a Privacy-first, offline-capable Pokemon card scanner that runs 100% in your browser using modern Web AI.

It combines YOLO11 for object detection and MobileCLIP for image embedding to identify cards instantly, without uploading any images to a server.

Features

  • Instant Detection: Uses YOLO11n to detect cards in real-time (~10ms)
  • Client-Side Recognition: identifies cards using MobileCLIP-S2 embeddings directly in the browser
  • Privacy First: No images are ever uploaded to a server
  • Offline Capable: Works fully offline after the initial load
  • Comprehensive Database: Merged data from TCGdex and LimitlessTCG for English and Japanese cards
  • Price Checking: Fetches real-time pricing from TCGplayer

Tech Stack

  • YOLO11n for object detection (via ONNX Runtime)
  • MobileCLIP-S2 for image embeddings
  • TensorFlow.js & ONNX Runtime Web for inference
  • WebGPU & WebAssembly for hardware acceleration
  • PocketBase for backend data sync (prices)
  • Vite & React for the frontend

Read the full story on how I built this

  • machine learning
  • web