Object Detection iOS 🔍

A high-performance, real-time object detection application for iOS, leveraging YOLOv3, CoreML, and Apple's Vision framework.

Features

• Live Detection: Real-time object recognition using the YOLOv3 model trained on the COCO dataset.
• Neural Acceleration: Explicitly optimized for the Apple Neural Engine (ANE) for maximum speed and battery efficiency.
• Glassmorphic UI: High-end SwiftUI overlay with real-time controls for confidence thresholds.
• Visual Feedback: Dynamic bounding boxes that scale perfectly with the camera's field of view.

Technical Focus

• Safety: Built with Swift 6 strict concurrency to ensure heavy ML inference never creates data races.
• Accuracy: Uses ContinuousClock to measure inference latency with nanosecond precision, providing transparent performance metrics.
• Throughput: Implements a "Smart Back-Pressure" system to prioritize the most recent camera frame, keeping the UI at 60+ FPS even during heavy detection.

Tech Stack

• Swift 6
• SwiftUI
• CoreML / Vision
• AVFoundation

State Management

• isProcessing: Prevents frame-stacking and ensures the pipeline stays reactive.
• confidenceThreshold: Real-time filtering of ML predictions via @Observable.
• inferenceTime: Live performance tracking for hardware benchmarking.

Requirements

• Xcode 16.0+ (Required for Swift 6 features)
• iOS 17.0+ (Required for modern @Observable and Vision performance)
• Physical Device: Required for Neural Engine (ANE) and camera testing.

License

Created by Jonni Åkesson.

Open for educational and portfolio use.

---