ollamaocr-python 0.1.0

ollamaocr with Llama vision

Ollama OCR

A Python-based OCR tool leveraging the Llama 3.2-Vision model for highly accurate text recognition from images, preserving original formatting and structure.

Features

• 🚀 High Accuracy: Text recognition powered by the Llama 3.2-Vision model.
• 📝 Preserves Formatting: Maintains the original structure and layout of the recognized text.
• 🖼️ Wide Format Support: Works with image formats such as .jpg, .jpeg, and .png.
• ⚡️ Customizable Output: Returns results in either Markdown or JSON format.
• 💪 Robust Error Handling: Ensures smooth processing with clear error messages for unsupported formats or invalid configurations.

---

System Requirements

• Python 3.8 or higher
• Ollama Server running locally
• Llama 3.2-Vision model installed

Prerequisites

1. Ensure the Ollama server is running before using the tool.
2. Download and configure the Llama 3.2-Vision model for OCR tasks.

---

Instalation

pip install ollamaocr-python

---

Usage

Basic Usage

from ollamaocr import OllamaOCR

# Initialize the OCR tool
ocr = OllamaOCR()

# Perform OCR in Markdown format
markdown_result = ocr.perform_ocr("path/to/image.jpg", output_format="markdown")
print(markdown_result)

# Perform OCR in JSON format
json_result = ocr.perform_ocr("path/to/image.jpg", output_format="json")
print(json_result)

Error Handling

The class provides comprehensive error handling for unsupported formats or invalid configurations:

from ollamaocr import OllamaOCR

ocr = OllamaOCR()

try:
    result = ocr.perform_ocr("invalid_file.bmp", output_format="markdown")
except ValueError as e:
    print(f"Error: {e}")

Customizable Prompts

Modify the prompts used for OCR to suit specific requirements:

• Markdown Prompt: Preserves formatting in Markdown structure.
• JSON Prompt: Outputs results in JSON format.

Limitations

Currently supports only .jpg, .jpeg, and .png image formats. Requires the Ollama server to be running locally with the Llama 3.2-Vision model installed.

License

This project is licensed under the MIT License.