How krishnazyala/facerecognition works
Project type and purpose
• This is a native Android mobile application whose primary function is real-time face detection and recognition using on-device ML models and a local database.
• The README confirms it’s built in Kotlin with Jetpack Compose for UI, ML Kit for detection, TensorFlow Lite (Mobile FaceNet) for recognition, and Room for persistence.
File structure
• Top-level “app” module contains:
• src/main/kotlin/com/krishnaZyala/faceRecognition/app – application entry point (MAIN.kt) and DI setup (Module.kt)
• data – Repository, Room database (MainDatabase.kt), DAO, type converters, and data models (AppState, FaceInfo, etc.)
• lib – utility wrappers around TFLite (AiModel), file/media utilities, logging, string utils
• ui – Jetpack Compose components:
– navigation (AppHost, Routes)
– composables (FaceView, FrameView, FaceAnalytics)
– screens with MVVM (HomeScreen, RecogniseFaceScreen, AddFaceScreen, etc.) each paired with a ViewModel
• assets – TFLite model files (.tflite)
• androidTest / test – instrumentation and unit tests
• Build and config files: Gradle scripts, AndroidManifest, ProGuard rules, resource XMLs.
README for additional insights
• Confirms features: real-time detection, name display, add/delete faces.
• Lists tech stack: Mobile FaceNet, ML Kit, TFLite, Jetpack Compose, Kotlin.
Main components and interactions
a. Components
• Camera Input (Android CameraX / system)
• Face Detection Engine (ML Kit)
• Face Recognition Engine (TensorFlow Lite via AiModel)
• Repository (Repository.kt)
• Local Database (Room: MainDatabase + MainDao)
• UI Layer (Jetpack Compose screens & composables)
• ViewModel Layer (per-screen ViewModel)
• DI Container (Module.kt)
• Assets (TFLite model files)
• Utility Library (FileUtils, MediaUtils, etc.)
b. Relationships & data flows
1. Camera → frames → Face Detection (ML Kit)
2. Face Detection → crop → Face Recognition (AiModel + .tflite)
3. Recognition result (embedding) → Repository
4. Repository ↔ Room Database (store/retrieve face embeddings & metadata)
5. ViewModel requests data from Repository; receives recognition results and/or stored faces
7. UI actions (Add/Delete) → ViewModel → Repository → Database
8. DI Container wires all dependencies (ViewModels, Repository, AiModel, Database) at application start (MAIN.kt)
c. Architectural patterns
• MVVM: separation of UI (Compose) from business logic (ViewModels) and data (Repository/Room)
• Dependency Injection: Module.kt defines singletons for AiModel, Repository, Database, etc.
• Modular layers: UI, domain (ViewModel/Repository), storage (Room), ML inference
d. Key technologies
• Jetpack Compose (UI)
• Android CameraX or Camera API
• ML Kit (face detection)
• TensorFlow Lite (recognition)
• Room (persistence)
• Kotlin Coroutines / Flow (async data streams)
Final checklist
• Clearly separate layers: Presentation (UI), ViewModel, Domain/Repository, Persistence, ML
• Interactions and data formats (camera frames, embeddings, metadata)
• Major libraries and frameworks
• Flow for both “recognize face” and “add/delete face” user journeys are captured