OfflineID

An offline employee attendance system that performs face recognition directly on-device using TensorFlow Lite and FaceNet, without requiring an internet connection.

Features

• Offline face recognition
• Face embedding generation using FaceNet
• Liveness detection before verification
• Automatic attendance logging
• Local SQLite database
• Local image storage
• Native Android TensorFlow Lite integration
• React Native cross-platform UI

---

Tech Stack

Frontend

• React Native
• TypeScript

Machine Learning

• TensorFlow Lite
• FaceNet
• Cosine Similarity

Storage

• SQLite
• Android File System

Native Android

• Kotlin
• Native Modules
• Vision Camera

---

Project Architecture

Camera
      │
      ▼
Vision Camera
      │
      ▼
Captured Face
      │
      ▼
TensorFlow Lite
      │
      ▼
FaceNet
      │
      ▼
128D Face Embedding
      │
      ▼
Cosine Similarity
      │
      ▼
SQLite Database
      │
      ▼
Attendance Recorded

---

Project Structure

src
│
├── components
├── database
├── repositories
├── screens
├── services
├── utils

---

How It Works

1. User completes liveness detection.
2. Camera captures a face image.
3. TensorFlow Lite runs the FaceNet model.
4. A face embedding is generated.
5. Cosine similarity compares the embedding against enrolled users.
6. If similarity exceeds the threshold, attendance is marked.
7. All processing happens locally on the device.

---

Future Improvements

• Cloud synchronization
• Admin dashboard
• Multi-device support
• Face anti-spoofing
• Employee analytics
• Real-time attendance monitoring

---