readgrid 0.2.2

A document layout pipeline for detecting tables, images, and structured extraction.

readgrid

readgrid is a Python package for document layout analysis and content extraction.
It lets you upload scanned documents, automatically detect tables and images, manually adjust bounding boxes for all document elements, and extract clean structured output using LLMs.

---

✨ Features

• Stage 1 – Upload & Detect

  • Upload document images in batch mode.
  • Automatically detect tables (red boxes) and images (green boxes).
  • Manually draw bounding boxes for columns (blue boxes), headers (orange boxes), and footers (magenta boxes).
  • Interactive editor supports multiple boxes per type with add/edit/delete functionality.

• Stage 2 – Coordinate Testing

  • Verify detected regions with side-by-side previews.
  • Test either existing detections or custom coordinates.
  • Works with all box types (tables, images, columns, headers, footers).

• Stage 3 – Content Extraction

  • Extract structured JSON output with layout-aware reading order.
  • Supports multi-column layouts (single, two, three, four columns).
  • Replace detected tables with clean HTML <table> tags.
  • Insert [image] placeholders with captions in correct reading order.
  • Proper header/footer separation from main content.
  • Supports LaTeX formatting for mathematical equations.
  • Customizable output fields with inclusion/exclusion options.

• Utility Functions

  • pretty_print_page_with_image() – inspect extracted results with annotated images.
  • show_comparison_view() – compare annotated vs. reconstructed content with multiple view modes.
  • cleanup_pipeline() – reset all artifacts.
  • editor() – interactive JSON correction with LLM assistance and diff visualization.

---

🚀 Installation

pip install readgrid

---

🛠️ Usage

Stage 1: Upload, Detect, and Edit

from readgrid import stage_1

stage_1()

Stage 2: Test Coordinates

from readgrid import stage_2

stage_2(
    row_id="ID_1",
    box_type="tables",
    box_index=0
)

Stage 3: Extract with Gemini

from readgrid import stage_3

stage_3(api_key="YOUR_API_KEY")

---

✏️ Editor

from readgrid import editor

# Basic usage - will prompt for API key
editor(row_id="ID_1")

# With all options
editor(
    row_id="ID_1",
    api_key="YOUR_API_KEY",
    model="gemini-2.0-flash",
    clean=True,        # Auto-clean tables, LaTeX, newlines
    font_size=12       # Adjust display size
)

Features:

• Currently supports just JSON file types.
• Automatic cleaning of tables, LaTeX formatting, and broken words
• Side-by-side view: original image vs. current JSON
• Conversational interface: "Add newline after Table 1" or "Fix the header"
• Real-time diff view (unified or side-by-side with color coding)
• Undo/reset functionality
• Saves corrections back to final_outputs/

---

📦 Requirements

• Python 3.8+
• OpenCV
• NumPy
• Pillow
• google-generativeai

---

📄 License

MIT License. See LICENSE for details.