Image Review Tool

A Python-based desktop application for reviewing annotated images. It supports multiple annotation formats (COCO, PASCAL VOC, YOLO, and WeedCOCO) and displays them over images in a practical three-pane layout. Built with Tkinter, it includes features to streamline the review process and analyze datasets effectively.

Features

• Annotation Format Support:
  Loads annotations from various formats:

  • COCO: JSON files with segmentation masks or bounding boxes.
  • PASCAL VOC: XML files with bounding box annotations.
  • YOLO: Text files with bounding boxes, dynamically reading image dimensions; supports an optional labels.txt for class names.
  • WeedCOCO: An extended COCO format with additional metadata like agricultural contexts, shown in a dedicated tab.

• Annotation Display:
  Overlays annotations—bounding boxes or segmentation masks—on images, depending on what’s provided. Highlights specific annotations when selected.

• Filtered Cutouts:
  Filters images by annotation class and displays cropped thumbnails of annotated regions. Thumbnails are cached for faster loading, and resizing is optimized to keep the interface responsive.

• Statistics and Heat Maps:

  • Bar charts summarize annotation counts per class, with interactive Matplotlib controls.
  • Heat maps show the spatial distribution of annotations across images, useful for spotting patterns.

• User Interface:

  • Left Panel: Lists annotations for the current image, with a filter option.
  • Center Panel: Displays the main image with navigation, zoom controls, and a persistent comment section.
  • Right Panel: Contains tabs:
    • Load Data: Set image/annotation directories, output file, and annotation type.
    • Filtered: View class-specific cutouts.
    • Comments: Add and review notes per image.
    • Stats: Generate annotation statistics.
    • Heat Map: Visualize annotation placement.
    • Metadata: Display WeedCOCO-specific details (e.g., agricultural context).

• Persistent Settings and Comments:
  Saves configuration (directories and output file) to a settings file in your home directory. Comments are stored in a temporary JSON file and persist across sessions, linked to the dataset.

• Export Functionality:
  Exports comments, annotation counts, and class details to an Excel file for further analysis.

Installation

Requirements

• Python 3.x
• Tkinter (typically included with Python)
• opencv-python
• Pillow
• pandas
• matplotlib
• numpy

Install Dependencies

pip install opencv-python Pillow pandas matplotlib numpy