Viso

Viso is a mobile information-measurement system (iOS app) for recognizing objects in images and generating human-readable explanations. It uses Apple’s Vision framework for on-device object detection and FoundationModels to generate GPT-style text descriptions — all running fully offline.

Demo

2025-10-11.17.42.14.mov

🧠 Features

• Load images from camera or photo library
• Detect multiple objects using the Vision framework
• Generate natural language explanations using FoundationModels
• Export results (object labels and confidence scores) to CSV
• Entirely offline — no internet required
• Built with The Composable Architecture (TCA)

📱 Interface Overview

• Main screen: View analyzed images and top predictions
• Result screen: Tap an image to view a generated explanation
• Export: One-tap CSV export with all object data
• Smooth navigation, iOS-native UX following Human Interface Guidelines

🛠 Technology Stack

• Swift 5.9 / Swift 6
• Vision Framework
• FoundationModels
• The Composable Architecture (TCA)
• SwiftUI & UIKit
• Local CSV export module

📂 Example CSV Output

ID,Observation,Confidence
E9D4732F...,table,0.87
E9D4732F...,vegetable,0.76
E9D4732F...,salad,0.72

Each row represents an object detected in an image with its confidence score.

📦 Installation (for developers)

1. Clone the repository:

   git clone https://github.com/jaroshevskii/Viso.git

2. Open the project in Xcode:

   open Viso.xcodeproj

3. Build and run on a physical iOS device (iOS 26+ recommended)

⚠️ Note: FoundationModels require Apple Silicon or A17+ for local inference

🧪 Sample Result

Given an image of a salad on a wooden table:

Detected objects:

• structure (0.93)
• wood_processed (0.93)
• food (0.76)
• salad (0.69)

Generated explanation:

"The image shows a wooden table with a vegetable-based dish, likely a salad, and accompanying kitchen utensils."

📃 License

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