OCR Annotation Tool

A fast, efficient web-based tool for manually labeling text in image crops for OCR training datasets.

Features

• Fast Image Loading: Optimized batch loading with preloading for smooth workflow
• Random Image Order: Images load in random order to prevent annotation bias
• Keyboard Shortcuts: Quick navigation with Enter to save, Ctrl+Enter to skip
• RTL Text Support: Proper support for Persian, Arabic, and other RTL languages
• Progress Tracking: Real-time statistics on completed, skipped, and remaining images
• Undo Functionality: Go back to previous images and modify annotations
• Default Text Integration: Support for OCR model predictions as starting points

Requirements

• Python 3.6 or higher
• Flask: pip install flask

Installation

1. Clone the repository:

git clone https://github.com/faezeam/ocr-annotation-tool
cd ocr-annotation-tool

2. Install dependencies:

pip install flask

3. Run the application:

python app.py

4. Open your browser and go to: http://127.0.0.1:5000

Usage

Setup

1. Place your image crops in the crops/ folder
2. (Optional) Create labels/default.txt with OCR predictions in in tab-separated format: filename.jpg predicted_text

Annotation Workflow

1. View Image: Each image displays in the center with filename shown above
2. Enter Text: Type the text you see in the input field (supports RTL languages)
3. Save: Press Enter to save annotation and move to next image
4. Skip: Press Ctrl+Enter or Alt+S to skip difficult images
5. Undo: Click "Previous" button to go back and modify previous annotations

Keyboard Shortcuts

• Enter - Save annotation and go to next image
• Ctrl+Enter or Alt+S - Skip current image
• Esc - Clear input field

File Organization

• crops/ - Place your image files here (.jpg, .png, .bmp, .tiff, .gif)
• completed/ - Successfully annotated images are moved here
• skipped/ - Skipped images are moved here
• labels/labels.txt - All annotations saved in tab-separated format
• labels/default.txt - Optional OCR predictions (tab-separated: filename→text)

Output Format

Annotations are saved in labels/labels.txt in tab-separated format