Pizza Check CV — Phase 2

Author: Sheri
Version: 2.0.0
Platform: macOS (Mac mini M2) · Electron + Python

A production-grade computer vision platform for real-time pizza quality control at the store edge. Phase 2 transforms the Phase 1 proof of concept into a deployable system that validates topping accuracy, cheese portioning, and sauce distribution against live order data.

---

What It Does

Pizza Check CV monitors the pizza make line through a Sipeed MaixSense A075V RGB-D camera and runs YOLOv11 inference to:

• detect 26 topping classes (including separate diced and sliced tomato detection)
• estimate cheese quantity and coverage within ±10–15%
• validate sauce presence, amount, and distribution
• compare the assembled pizza against the active POS order in real time
• emit pass / warning / fail outcomes for each monitored build
• surface results through an Electron desktop GUI designed for store operators

When POS integration is unavailable, the system falls back gracefully to recipe-only validation mode without crashing.

---

Project Structure

pizza-check-cv/
├── src/
│   └── pizza_check_cv/
│       └── backend/
│           └── app.py          # FastAPI edge backend
├── electron/
│   ├── main.js                 # Electron shell & backend bridge
│   ├── preload.js              # Safe renderer bridge
│   └── renderer/               # Production monitoring UI
├── scripts/
│   └── run_backend.py          # Backend launcher
├── data/
│   ├── config.json             # Local configuration & recipe seed
│   └── orders.json             # Local order queue seed
├── tests/
│   └── test_backend_api.py     # Backend smoke coverage
├── package.json                # Electron / Node dependencies
├── pyproject.toml              # Python package & dependencies
└── yolo11n.pt                  # Default YOLOv11 nano weights

---

Getting Started

Prerequisites

• Python 3.13+
• Node.js 20+
• (Optional) Sipeed MaixSense A075V RGB-D camera at 192.168.233.1

Python Setup

pip install -e ".[dev]"

Run the Backend

python scripts/run_backend.py

The backend starts at http://127.0.0.1:8765 and exposes:

Method	Endpoint	Description
GET	/health	Service health check
GET	/api/state	Current pipeline state
GET	/api/config	Active configuration
PUT	/api/config	Update configuration
GET	/api/orders	Order queue
POST	/api/orders	Inject an order
POST	/api/integration/mode	Set integration mode
GET	/api/recent-detections	Latest CV detections
GET	/api/logs	Structured log stream
POST	/api/simulate	Simulate a detection event
POST	/api/capture	Trigger a manual capture
WS	/ws	Real-time event stream

Run the Electron App

npm install
npm start

The launcher automatically sets PYTHONPATH=src so the desktop app boots against the in-repo Python service without a separate package install.

Run Tests

python -m unittest discover -s tests -v

---

Camera Integration

The production camera is the Sipeed MaixSense A075V RGB-D sensor, accessed over HTTP at the vendor default address 192.168.233.1:80.

The backend sends frame configuration to /set_cfg and fetches RGB-D frames from /getdeep. Each frame payload contains a 16-byte header, a 12-byte config block, then depth, IR, status, and optional RGB data. RGB frames arrive as JPEG and are decoded to raw bytes before entering the CV pipeline.

A synthetic fallback is included for development and testing without hardware.

---

CV Pipeline & Model Weights

YOLOv11 handles all visual inference tasks:

Model slot	Default	Purpose
pizzaWeights	yolo11n.pt	Pizza presence & format detection
toppingWeights	(supply your 26-class model)	Topping identity & counting
qualityWeights	(supply your quality model)	Sauce / cheese heuristics

Store custom weights under models/yolo11/ and point to them via data/config.json or an environment variable. When optional weights are absent the backend starts in degraded mode — the API remains functional and reports pipeline health accurately.

---

Phase 2 Objectives

Goal	Target
Topping classes	26 (YOLO-based)
Cheese quantity accuracy	±10–15%
End-to-end validation latency	≤ 2 seconds
Scheduled uptime (pilot)	≥ 99.5%
Sauce / topping precision & recall	≥ 90%
Cheese precision & recall	≥ 88%

Full requirements and acceptance criteria are documented in PHASE_2_PIZZA_CHECK_CV_SPEC.md.

---

Tech Stack

Layer	Technology
CV inference	YOLOv11 (Ultralytics)
Camera	Sipeed MaixSense A075V RGB-D
Backend	Python 3.13 · FastAPI · Uvicorn
Desktop shell	Electron 32
Real-time comms	WebSocket (ws)
Image processing	OpenCV (headless)
HTTP client	Requests
Test framework	pytest / unittest

---

License

Private — all rights reserved.

---

Pizza Check CV Phase 2 · Authored by Sheri