Browser ONNX OCR Lab

A browser-only OCR experiment lab that compares preprocessing pipelines in real time.

MVP implemented

• Vite + TypeScript web app
• Image upload and 3 pipeline comparison:
  • raw
  • grayscale
  • binary
• Per-pipeline output card with:
  • transformed image preview
  • prediction text (6-digit)
  • confidence score
  • latency (ms)
• Test suite for preprocessing and inference wrapper logic

Why this is portfolio-worthy

• Shows practical ML product thinking: pipeline quality is treated as measurable, not cosmetic.
• Demonstrates browser-side inference workflow shape (no backend dependency).
• Provides clear extension point to plug in a real ONNX OCR model.

Quick start

npm install
npm run dev
npm test
npm run build

Current architecture

• src/preprocess.ts: image preprocessing stages
• src/ocr.ts: inference wrapper (MVP uses deterministic placeholder path)
• src/main.ts: UI orchestration and pipeline comparison rendering
• tests/: unit tests for preprocessing + inference wrapper

Next steps

1. Replace placeholder inference with real ONNX model session loading
2. Add character-level confidence visualization
3. Add sample dataset replay and CSV benchmark export