OptiFrame YCrunch

Active development has moved to George-Fam/optiframe. This repository is the CodeML 2026 hackathon submission and is no longer updated.

From recycled eyeglass lenses to a 3D-printed frame. CodeML 2026 Challenge, Santé Numérique Sans Frontières.

• Application: https://optiframe.app
• Target accuracy: 0.5 mm for A, B, and the bridge, according to ISO 12870. The jury awards maximum points at 1 mm. See Validation.

flowchart LR
    A[/"Photo of a lens<br/>on the sheet"/] --> B["Measurements in mm<br/>A, B, perimeter"]
    B --> C(["1:1 SVG outline<br/>for paper verification"])
    B --> D["3D frame<br/>for two lenses"]
    D --> E(["frame.stl<br/>ready to print"])

    classDef result fill:#dcfce7,stroke:#15803d,color:#0b3d1c
    class C,E result

Loading

Architecture

Three parts: the application on the phone, the server that performs the measurements, and the AI model training, which is done once in advance.

flowchart TB
    subgraph PHONE["Phone: web application (Angular)"]
        CAM[/"Camera or photo import"/]
        CARDS["Lens card<br/>measurements + control image"]
        GEN["Frame generator<br/>(manifold-3d)"]
        VIEW["3D preview<br/>(three.js)"]
        EXP(["Downloads<br/>frame.stl, outline.svg"])
    end

    subgraph SERVER["Server: API (Spring Boot)"]
        RECT["Correction<br/>ArUco markers (OpenCV)"]
        SEG["Lens detection<br/>AI model or classical method"]
        MEAS["Measurement<br/>contour in mm, A, B, perimeter"]
    end

    subgraph TRAIN["Training (Python, Colab)"]
        DATA[("Dataset<br/>photos + masks")]
        MODEL[("U-Net model<br/>exported to ONNX")]
    end

    SHEET[("Reference sheet<br/>sheet-layout.json")]

    CAM -- "photo" --> RECT
    RECT --> SEG --> MEAS
    MEAS -- "contour in mm" --> CARDS
    CARDS -- "right lens + left lens" --> GEN
    GEN --> VIEW
    GEN --> EXP
    MEAS -. "auto-labeled photos" .-> DATA
    DATA --> MODEL
    MODEL -. "lens-seg.onnx" .-> SEG
    SHEET -. "marker positions" .-> RECT

    classDef phone fill:#dbeafe,stroke:#1d4ed8,color:#0b1f4d
    classDef server fill:#dcfce7,stroke:#15803d,color:#0b3d1c
    classDef train fill:#ffedd5,stroke:#c2410c,color:#431407
    class CAM,CARDS,GEN,VIEW,EXP phone
    class RECT,SEG,MEAS server
    class DATA,MODEL train

Loading

Blue: on the phone. Green: on the server. Orange: prepared in advance. Cylinders represent files or data.

Folder	Contents
frontend/	Angular 22 application (PWA)
backend/	Spring Boot 4 API, POST /api/measure
training/	Reference sheet, dataset, training, ONNX export

Documentation

Page	Contents
How it works	User workflow, photo checks, control images
Frame	Frame generation, keeping the lens inside the circle, conventions
Capture	Reference sheet, lighting, photography
Data and AI	Auto-labeled collection, training, tools, and licenses
Validation	Measured deviations, known limitations, and mitigations
Deployment	Cloud Run, Cloud Build, environment variables

Run locally

# API (Java 25+)
cd backend && ./mvnw spring-boot:run          # http://localhost:8080

# App (Node 22.22+, 24.15+ or 26+)
cd frontend && npm install && npm start       # http://localhost:4200

# Reference sheet
cd training && pip install -r requirements.txt && python make_sheet.py

# Test on a phone (the camera requires HTTPS)
cloudflared tunnel --url http://localhost:4200

Tests: cd backend && ./mvnw test (end-to-end measurement on a synthetic tilted photo) and cd frontend && npm test.

Formatting and continuous integration

Code	Tool	Fix
TypeScript, HTML, CSS, JSON, YAML, Markdown	Prettier, ESLint (Angular)	cd frontend && npm run format && npm run lint -- --fix
Java, pom.xml	Spotless (Eclipse)	cd backend && ./mvnw spotless:apply
Python	Ruff	ruff format . && ruff check --fix .
Shell	shfmt	shfmt -w deploy/*.sh .githooks/*

• Running npm install in frontend/ activates the .githooks/pre-commit hook: staged files are formatted before each commit (Ruff and shfmt must be installed for Python and shell files).
• GitHub Actions (ci.yml) checks formatting, linting, tests, and the build on every push and pull request. Deployment remains handled by Cloud Build on main.

License

MIT. The libraries used retain their own licenses; see Data and AI.