image2md

Structured Images to Markdown Converter

This tool automatically converts batch of images containing structured data (tables, formulas, graphs, diagrams, flowcharts, etc.) into markdown format. Markdown files suitable for RAG pipeline. It uses Anthropic's models via API to analyze images and create detailed markdown descriptions based on included robust system prompt.

Prerequisites

Before you start, you need to have:

1. Python installed on your computer (version 3.7 or higher)
2. An Anthropic API key (get it from Anthropic's console)

Installation Steps

1. Install Python

If you don't have Python installed:

1. Go to Python's official website
2. Download the latest version for your operating system
3. Run the installer
   • On Windows: Make sure to check "Add Python to PATH" during installation
   • On Mac: Follow the standard installation process

2. Get the Code

Using Git (Option 1):

1. Open Terminal (Mac) or Command Prompt (Windows)
2. Navigate to where you want to save the project:

cd Documents

3. Clone the repository:

git clone https://github.com/PetrAPConsulting/image2md.git

4. It creates folder image2md with cloned files

Manual Download (Option 2):

1. Click the green "Code" button on this page
2. Click on the sheet "Local"
3. Select "Download ZIP"
4. Extract the ZIP file to your desired location

3. Install Required Library

Open Terminal (Mac) or Command Prompt (Windows) in the project folder and run:

pip install anthropic

If that doesn't work, try:

pip3 install anthropic

Configuration

1. Open the images.py file in a text editor
2. Find this line:

self.client = anthropic.Anthropic(api_key="insert_api_key_here")

3. Replace "insert_api_key_here" with your Anthropic API key
4. Follow development of Anthropic models and make adjustments in the script when new version is realised. Only models with vision capabilities are supported.

def __init__(self, model: str = "claude-3-5-sonnet-20241022")

def main():
    available_models = [
        "claude-3-5-sonnet-20241022",
        "claude-3-opus-20240229",
        "claude-3-haiku-20240307"
    ]  

Usage

1. Copy your images (.jpg, .jpeg, or .png) to the same folder as the script. Keep images around 1000px x 1000px for token consumption optimalization
2. Open Terminal (Mac) or Command Prompt (Windows)
3. Navigate to the script's folder:

cd path/to/your/folder

4. Run the script:

python images.py

5. Select a model when prompted (1-3)
6. The script will create markdown (.md) files for each image in the same folder

Supported File Types

• .jpg
• .jpeg
• .png

Features

• Automatically detects tables, formulas, graphs, flowcharts, etc.
• Creates markdown tables from image tables
• Converts mathematical formulas to LaTeX format
• Provides detailed analysis of graphs with key values
• Creates nice clear markdown mermaid from flowcharts and process diagrams
• Preserves anotations and tables with measurements
• Generates log files for troubleshooting
• IMPORTANT: If you need output in different language than ENG you need to translate system prompt in python script. Even though Anthropic models are multilingual, language of sytem prompt determinates language of output except data it directly transfers from source, like content of tables.

Troubleshooting

Common Issues:

1. "No module named 'anthropic'"

   • Run pip install anthropic again
   • Make sure you're using the correct Python version

2. "Invalid API key"

   • Check if you've correctly inserted your API key in the script
   • Verify your API key is active on Anthropic's website

3. "Python not found"

   • Make sure Python is installed
   • Try using python3 instead of python

Still Having Problems?

First, try your good friend ChatGPT, Claude or Gemini. All three of them are able to help you if you give them occured errors. Or create an issue in this repository with:

• The error message you're seeing
• Your operating system
• Steps you've tried

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

• Uses Anthropic's API for image analysis
• Inspired by Petr's need for automated structured data extraction from images