YOLO Object Detection Demo

A real-time object detection app for iOS using YOLO (You Only Look Once) and CoreML. This app demonstrates object detection using the camera feed with bounding boxes and confidence scores.

Features

• Real-time object detection using YOLO model
• Camera preview with live detection boxes
• Confidence scores for each detection
• Non-Maximum Suppression (NMS) to eliminate overlapping detections
• Configurable parameters:
  • Confidence threshold
  • IoU threshold for NMS
  • Minimum and maximum box sizes
• Support for 80 COCO classes
• Debug logging for coordinate transformations

Requirements

• iOS 15.0+
• Xcode 13.0+
• Swift 5.5+
• CoreML
• Vision framework

Installation

1. Clone the repository
2. Open yolodemo.xcodeproj in Xcode
3. Build and run on a compatible iOS device

Usage

1. Launch the app
2. Grant camera permissions when prompted
3. Point the camera at objects to detect
4. Adjust detection parameters using the sliders:
   • Confidence threshold (0.1-1.0)
   • IoU threshold (0.1-1.0)
   • Min box size (0-100)
   • Max box size (100-1000)

Technical Details

Model

• Uses YOLO model converted to CoreML format
• Input size: 640x640
• Output: Bounding boxes with confidence scores for 80 COCO classes

Coordinate Handling

• Properly handles aspect ratio preservation
• Scales coordinates from model space (640x640) to screen space
• Accounts for image scaling and centering offsets

Non-Maximum Suppression

• Eliminates overlapping detections
• Uses IoU (Intersection over Union) threshold
• Preserves highest confidence detection when overlaps occur

Debug Output

The app includes detailed debug logging for:

• Model loading and configuration
• Coordinate transformations
• Detection processing
• NMS results

License

This project is available under the MIT license. See the LICENSE file for more info.