ocrmypdf-easyocr 0.3.0

Plugin to run OCRmyPDF with the EasyOCR engine

OCRmyPDF EasyOCR

This is a plugin to run OCRmyPDF with the EasyOCR engine instead of Tesseract OCR, the default OCR engine for OCRmyPDF. Since EasyOCR is based on PyTorch, it makes use of Nvidia GPUs. Hopefully it will be more performant and accurate than Tesseract OCR.

It is currently experimental and does not implement all of the features of OCRmyPDF with Tesseract, and still relies on Tesseract for certain operations.

Requirements

• OCRmyPDF 17 or newer. As of 0.3.0, this plugin returns structured OCR results to OCRmyPDF, which renders the text layer itself using its built-in renderer. (For OCRmyPDF 14–16, use a < 0.3.0 release of this plugin.)
• Python 3.11 or newer.

Installation

To use this plugin, first install PyTorch according to the official instructions, which may differ for your platform.

Then install OCRmyPDF-EasyOCR into the same virtual environment or conda environment as you installed PyTorch:

pip install git+https://github.com/ocrmypdf/OCRmyPDF-EasyOCR.git

Once installed, the plugin is discovered automatically through its entry point, and OCRmyPDF uses EasyOCR for OCR by default (i.e. for --ocr-engine auto). You can also select it explicitly with --ocr-engine easyocr, or fall back to the built-in engines with --ocr-engine tesseract / --ocr-engine none. OCR still depends on Tesseract for some non-OCR tasks (see below).

Development

This project uses uv for package management.

uv sync --group test   # create the virtual environment and install dependencies
uv run pytest          # run the test suite

The tests need poppler's pdftotext and a Tesseract installation available on PATH, and will download EasyOCR's models on first run.

Troubleshooting

If you see a log message Neither CUDA nor MPS are available - defaulting to CPU. Note: This module is much faster with a GPU then PyTorch is not installed with GPU support.

Pass --easyocr-no-gpu to force CPU mode, or --easyocr-batch-size to tune the EasyOCR batch size.

To do

Contributions, especially pull requests, are quite welcome!

This plugin is experimental and is missing some features:

• Tesseract is still required to determine page orientation.
• EasyOCR is effectively single-threaded, to eliminate race conditions; OCRmyPDF still parallelizes other stages of the pipeline across pages.