PaddleOCR

Package

Open-source OCR toolkit you self-host, with formula-to-LaTeX and PDF parsing

Price: Free, open source
Access: None, runs locally

About

Apache-2.0 OCR toolkit from the PaddlePaddle team that runs on your own machines. PP-OCRv6 reads 50 languages with one model (100+ in all), a formula pipeline turns equation images into LaTeX, and PP-StructureV3 and PaddleOCR-VL convert PDFs to Markdown or JSON. Needs PaddlePaddle.

What you can do with it

Convert images of math formulas into LaTeX on your own server
Read text from photos and scans in 100+ languages without a cloud API
Turn PDFs into Markdown or JSON with tables and formulas for RAG

Get started

Install PaddlePaddle for your hardware, then pip install paddleocr
Run paddleocr formula_recognition_pipeline -i equation.png

Example

from paddleocr import FormulaRecognitionPipeline

pipeline = FormulaRecognitionPipeline()
for res in pipeline.predict("./general_formula_recognition_001.png"):
    res.print()  # LaTeX for each recognized formula
    res.save_to_json(save_path="output")

Details

Hosting: Self-hosted, Runs locally
Available in: Worldwide
Official SDKs: Python, JavaScript/TypeScript
MCP server: Local