Manchu Unicode HTR for woodblock printing and handwritten Manchu documents from the Qing Period
Year: 2026
Type: dataset
Abstract: Manchu unicode HTR model for manchu This is a text recognition model trained on 242 transcribed pages (~3000 lines) from five manchu documents (xylographic and handwritten) made during the Qing period (1644-1911). Those pages were transcribed using Möllendorf romanisation and were converted into Manchu unicode using a Python script, hence giving the opportunity to train a dedicated model for transcribing directly in Manchu. This model was trained during the ManjugisunTranscript Project funded by the BnF Datalab and Huma-Num (2025-2026). The model transcribe manchu script into manchu Unicode. Punctuations marks are rendered using the latin characters. List of documents mandchou 117 mandchou 243 mandchou 289 mandchou 168 mandchou 268 Architecture The default model architecture and hyperparameters of kraken 6.0 where used. Uses The model is trained on a xylographic documents as well as well written manuscripts. It can be used as a base model for fine-tuning more specific models from it. How to Get Started with the Model Follow the instructions on installing and using kraken from the website. Metrics CER: 2%
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7213428101