ViNET

ViNET is an offline, real-time object detection and depth estimation iOS app designed to empower visually impaired users with instant, multi-modal feedback. By combining lightweight CoreML models with a minimalist interface, ViNET delivers object type, position, and precise distance information—via speech and haptic cues—without ever requiring a network connection.

---

Key Features

• Real-Time Object Detection

  • Detects 80 object categories (e.g., pedestrian, vehicle, obstacle) entirely on-device with YOLOv8s.

• Accurate Depth Estimation

  • Estimates meter-level distances using MiDaS_Small and CoreML optimizations.

• Multi-Modal Feedback

  • Speech: “A table is 75.4 centimeters to the right.”

• Offline & Privacy-First

  • No internet needed; all video frames and model weights stay on the user’s device

• Battery-Friendly

  • Maintains ~30 FPS preview while keeping CPU usage below 70%

---

Installation

1. Clone this repository:

   git clone https://github.com/your-org/vinet.git](https://github.com/xddlzx/ViNET.git

2. Open ViNET.xcodeproj in Xcode (14.0+).

3. Ensure you have an iPhone device (iOS 15.0+) connected or select a compatible simulator.

4. Build & run the app:

   • Product → Run (⌘R)

---

Usage

1. Grant Camera Access when prompted.
2. Tap the Start button to begin live object detection.
3. Listen for spoken cues.

---

Project Structure

ViNET/
├── Models/
│   ├── YOLOv8s.mlmodel
│   └── MiDaS_Small.mlmodel
├── Sources/
│   ├── CameraSession.swift
│   ├── ObjectDetector.swift
│   ├── DepthEstimator.swift
│   └── VoiceReporter.swift
├── Resources/
│   ├── LaunchScreen.storyboard
│   └── Assets.xcassets
└── ViNET.xcodeproj

---

Contributing

1. Fork the repo and create a new branch for your feature/bugfix.
2. Implement your changes and write unit tests where applicable.
3. Open a Pull Request against main and describe your changes.

---

License

This project is licensed under the MIT License. See LICENSE for details.