Dataset · CATMuS
Medieval Manuscripts for Text Recognition (HTR)
CATMuS/medieval
Images of medieval manuscripts and their text transcriptions in six European languages.
What's inside
The dataset contains 100 thousand to 1 million pairs of manuscript images and corresponding machine-readable text. The sample features materials in French, English, Dutch, Italian, Spanish, and Catalan. The project was created to standardize ground-truth annotation for recognizing complex historical scripts.
What you can do with it
- Training handwritten text recognition (HTR and OCR) models
- Fine-tuning Vision-Language Models (VLMs) for working with historical documents
- Evaluating archive digitization quality across different European languages
- Tasks
- Image → text
- Languages
- French, English, Dutch, Italian, Spanish, Catalan
- Size
- 100K–1M rows
- Downloads this month
- 3K
- Likes
- 31
Models trained on it
- medieval-data/trocr-medieval-baseImage → text 1.5K
- mradermacher/Medusa0.1Line-9B-GGUF 308
- mradermacher/Medusa0.1Line-4B-GGUF 269
- mradermacher/qwen2-vl-2b-catmus-GGUF 255
- ENC-PSL/Medusa0.2Line-4BImage → text 86
- ENC-PSL/Medusa0.2Line-9BImage → text 56
- medieval-data/trocr-medieval-latin-carolineImage → text 40
- ENC-PSL/Medusa0.1Line-4BImage → text 31
How to load it
from datasets import load_dataset
ds = load_dataset("CATMuS/medieval")