Structure

Top-level layout

ARShadesPDKit/
├─ Extensions/              # Math & SceneKit helpers (SCNVector, SCNMatrix)
├─ Services/                # Core services: ImageClassifierService, ResultBundle
├─ TfliteModel/             # TFLite model bindings (EfficientNet-lite variants)
├─ ViewModels/              # Business logic layer (MetricsViewModel)
├─ Views/                   # FacialMetricsViewController (PD measurement UI)
├─ Sources/
│   ├─ Constants/           # TFLite model assets (.tflite)
│   ├─ Models/              # Domain models (Consumption, Client)
│   ├─ Repository/          # Read/Write repositories for persistence
│   └─ ReadClient/          # Firebase/Firestore client (optional extension)

---

Core Components

FacialMetricsViewController

• Main entry point for PD measurement.
• Sets up ARSCNView and runs ARFaceTrackingConfiguration.
• Tracks eye transforms via ARKit and draws 3D nodes.
• Collects frames, applies classifier, and computes PD in millimeters.
• Provides real-time feedback:
  • “Come closer” if face is too far.
  • “Rotate your face” if yaw/roll/pitch is misaligned.
  • “Remove glasses” if classifier detects sunglasses.

MetricsViewModel

• Handles PD business logic and manages collected measurement values.
• Interfaces between FacialMetricsViewController and data consumers.

ImageClassifierService

• Wraps TFLite image classifier.
• Detects glasses vs no glasses using EfficientNet Lite models (efficientnet_lite0.tflite, efficientnet_lite2.tflite).
• Supports configurable thresholds (score, max results).

TfliteModel

• Enum wrapper for available TFLite models.
• Defines supported classification backbones (EfficientNet).

Repository Layer

• ReadWriteConsumption, ReadClient — designed to persist or sync measurements (Firebase or local).
• Optional: apps can integrate with Firestore or custom DBs.

Extensions

• Utility helpers for SCNVector3 and SCNMatrix4 operations.
• Used for 3D math: cross products, matrix multiplications, coordinate transforms.

---