A modern annotation tool for annotating visual relationships in COCO format.

# Install dependencies
pip install -r requirements.txt

# Run
python main_coco_pyqt.py --output name_of_output_file.json

# Run tkinter version
python main_coco.py --output name_of_output_file.json

• 📝 COCO Format Support: Load existing COCO JSON annotations
• 🔗 Relationship Annotation: Create subject-predicate-object relationships
• 🎯 Object Selection: Click or list-based object selection
• 💾 Multiple Export Formats: COCO, JSON, and TXT outputs
• ⌨️ Keyboard Shortcuts: Efficient workflow navigation

• 🌙 Modern Dark Theme: Nord color scheme
• 🎨 Custom Styled Components: Professional UI elements
• 🖼️ Advanced Canvas: Smooth image rendering with bounding boxes
• 📱 Responsive Layout: Splitter-based resizable panels
• 🎮 Enhanced Dialogs: Modern predicate selection

• 🔄 Fallback Compatibility: Works without PyQt6
• 📋 Dialog System: Custom dialog components
• 🖱️ Advanced Selection: Shift+click unselection support
• 📏 4K Display Support: Adaptive dialog sizing

• Python 3.7+
• PIL/Pillow for image processing

pip install -r requirements.txt

• It is recommended to annotate the bounding boxes with an external tool such as https://roboflow.com/. Roboflow is more practical for bounding box annotations and will save time with big datasets. You can also annotate bounding boxes with SGG-Annotate but it will be less efficient.
• Once you have created your dataset, export bounding box annotations in COCO format (json file).

• Click "Load COCO JSON" and select your COCO annotations file
• Select the folder containing your images

• Click "Load Relationships" and select a text file with relationship types
• File should contain one relationship per line (e.g., "on", "under", "next to")

• Click "Create Relationship" to enter annotation mode
• Select subject object (highlighted in blue)
• Select object object (highlighted in green)
• Choose predicate from the dialog

• Click "Save Relations" to export in multiple formats
• Files saved to output_coco_relations/ directory

Enhanced COCO JSON with rel_annotations and rel_categories fields:

{
  "rel_annotations": [
    {
      "id": 0,
      "subject_id": 123,
      "predicate_id": 1,
      "object_id": 456,
      "image_id": 789
    }
  ],
  "rel_categories": [
    {"id": 0, "name": "on"},
    {"id": 1, "name": "under"}
  ]
}

This format of annotations can be use with the SGG-Benchmark codebase to train SOTA SGG models.

# Modern modular approach
from pyqt6_ui.main_window import COCORelationAnnotatorPyQt
from pyqt6_ui.components import ModernButton, ModernLabel
from pyqt6_ui.canvas import ImageCanvas
from pyqt6_ui.dialogs import PredicateSelectionDialog

# Falls back to tkinter version
python main_coco.py

• Ensure PIL/Pillow is installed: pip install Pillow
• Check image file formats (JPG, PNG supported)

• Tool automatically scales images for display
• Original coordinates preserved for annotations

1. Fork the repository
2. Create feature branch (git checkout -b feature/AmazingFeature)
3. Commit changes (git commit -m 'Add AmazingFeature')
4. Push to branch (git push origin feature/AmazingFeature)
5. Open Pull Request