fgho/trocr-hanseXVII-kurrent

TrOCR Kurrent-Model Hanse 17th century

Handwritten Text Recognition model for administrative records from the northern German region, 17th century.

Model Description

• —Developed by: Research Centre for Hanse and Baltic History (FGHO, https://fgho.eu) in the Flow Project (https://flow-project.net)
• —Model type: TrOCR
• —Language(s) (NLP): Early New High German
• —Finetuned from model: dh-unibe/trocr-kurrent-XVI-XVII

Training Details

Training Data

The model was trained on administrative records from the northern German region dating to the 17th century, including:

• —Minutes of the Low German town assemblies (Niederdeutsche Städtetage; various archives)
• —Records of the maritime court (Seegerichtsakten; Lübeck City Archives)
• —Council minutes (Ratsprotokolle; Lübeck City Archives)

Epochs: 9.19/10 Best Eval CER: 0.0040 CER on test set: 0.0382