Downloads · 30 days
26
22% of all-time downloads
TinhinaneO/trocr-medieval-french-2026
trocr-medieval-french-2026 is a machine learning model from TinhinaneO. Use it for the machine learning task on the model card, and read the license before you ship it in a product. The card lists the license as mit.
Fine-tuning de microsoft/trocr-base-handwritten sur des manuscrits médiévaux en ancien et moyen français (XIIIe – XVe siècle).
Downloads · 30 days
26
22% of all-time downloads
All-time downloads
116
Public
Parameters
334M
1.3 GB on disk
Likes
0
Public
Click a slice to open those files.
.safetensors1.3 GB · 100%
From the Hugging Face model README
Fine-tuning de microsoft/trocr-base-handwritten sur des manuscrits
médiévaux en ancien et moyen français (XIIIe – XVe siècle).
Projet HETIC Master Data/IA — Module Vision par ordinateur 2026.
| Métrique | Baseline | Notre modèle |
|---|---|---|
| CER | 13.77% | 5.62% |
| WER | 42.41% | 15.46% |
IC Bootstrap 95% : [4.83%, 6.48%] sur 5 419 lignes
from transformers import TrOCRProcessor, VisionEncoderDecoderModel
from PIL import Image
processor = TrOCRProcessor.from_pretrained(
"TinhinaneO/trocr-medieval-french-2026"
)
model = VisionEncoderDecoderModel.from_pretrained(
"TinhinaneO/trocr-medieval-french-2026"
)
image = Image.open("ligne_manuscrit.jpg").convert("RGB")
pixel_values = processor(image, return_tensors="pt").pixel_values
generated_ids = model.generate(pixel_values, max_new_tokens=128)
texte = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
print(texte)
| Corpus | Lignes | Période | Licence |
|---|---|---|---|
| CREMMA Medieval | 45 696 | XIIIe-XVe s. | CC-BY |
| GalliCorpora XVe | 2 070 | XVe s. | CC-BY |
| Fabliaux | 11 874 | XIIIe-XIVe s. | CC-BY |