OCR-D wrapper for the party text recognizer

ocrd_party is an OCR‑D compliant workspace processor that wraps the Party text‐recognizer to perform line‐level OCR on existing baselines in PAGE‑XML documents.

party GitHub project: https://github.com/mittagessen/party

Installation

git clone https://github.com/mittagessen/party.git <path-to>/party
cd <path-to>/party
pip install .

cd <path-to>/ocrd_party
pip install -r requirements.txt
pip install .

Or install via Docker:

- docker compose build
- docker-compose run ocrd-party

For CPU only:

- docker compose build ocrd-party-cpu
- docker-compose run ocrd-party-cpu

Make sure your model directory contains a model.safetensors file compatible with Party.

Quick Start

Assuming you have already run layout and line segmentation (e.g. via ocrd-kraken-segment), recognize all lines in your workspace:

ocrd-party-recognize \
  -I OCR-D-SEG -O OCR-D-TXT \
  -P model /path/to/party/model \
  -P device cuda \
  -P batch_size 8

Parameters

model
Directory containing model.safetensors.

device
Compute device for inference (cpu, cuda:0, etc.).

batch_size
Number of lines to process in parallel.

Remark

ChatGPT 5 as well as Claude Opus 4.5 have been used to generate this OCR-D extension. At the beginning, the kraken extension was used as an example: github.com/OCR-D/ocrd_kraken Generated code has been manually tested and iteratively improved using the listed models.