TR - EN

Auto Image Labeler - User Guide

Auto Image Labeler is an application that performs object detection on images, simplifies labeling processes, and helps create datasets in YOLO format.

Table of Contents

• Installation
• Main Screen
• Image Loading and Navigation
• Manual Labeling
• Automatic Labeling
• Saving Annotations
• Dataset Splitting

Installation

1. Install required dependencies:

   pip install -r requirements.txt

2. Launch the application:

   python main.py

Main Screen

The application consists of the following sections:

• Image Area: Main area used for viewing and labeling images.
• Control Panel: Control buttons and settings located on the right side.
  • Navigation Buttons (Previous/Next)
  • Open Folder
  • Save Annotations
  • Clear Annotations
  • Class Management
  • Model Management
  • Dataset Management
  • Format Selection

Image Loading and Navigation

1. Open Folder: Click the "Open Folder" button to select a folder containing image files.
2. Navigate Between Images:
   • Use "Next" and "Previous" buttons to navigate between images.
   • Alternatively, use right/left arrow keys.

Manual Labeling

1. Drawing Rectangles:

   • Click and drag with the mouse on the image to draw a new rectangle.
   • Must be at least 5x5 pixels in size.

2. Editing Rectangles:

   • Click on an existing rectangle to select it.
   • Drag the selected rectangle with the mouse to move it.
   • Resize by dragging from edges and corners.
   • Delete the selected rectangle with the DELETE key.

3. Class Management:

   • Select an existing class from the "Classes" section.
   • Add new classes with the "Add Class" button.
   • Change the class of the selected rectangle with the "Set Class to Selected" button.

Automatic Labeling

1. Loading Model:

   • Click "Open folder" button in the "Model Management" section to select a YOLOv8 model (.pt).
   • Adjust the confidence threshold (default: 0.5).

2. Labeling:

   • Single Image: Click "Auto label current image" button to perform automatic object detection on the displayed image.
   • Batch Processing: Use "Simple Auto label all image" button to automatically label all images in the folder.

Saving Annotations

1. Format Selection:

   • Select your desired format from the "Output Format" section:
     • YOLO Format (class_id, x_center, y_center, width, height) - normalized
     • Standard Format (x, y, width, height, class_id) - pixel values

2. Saving:

   • Click "Save Annotations" button to save the annotations.
   • Annotations are saved to a subfolder named "annotations" in the folder containing the images.
   • Class information is stored in "classes.json" file.

Dataset Splitting

You can split your dataset into train, validation, and test sets for YOLO training:

1. Setting Ratios:

   • Set percentage ratios for train, validation, and test from the "Data splitter" section.
   • The sum of ratios must equal 100.

2. Splitting:

   • Click the "Split dataset" button.
   • Select the output folder.
   • When the process is complete, it automatically:
     • Creates train, validation, and test subfolders.
     • Creates data.yaml file.
     • Packages the entire dataset as a zip file.

Shortcuts

• Right/Left Arrow: Navigate to next/previous image
• DELETE: Delete selected