Downloads · 30 days
44
6% of all-time downloads
trocr_russian_18th_century_printed
dsmchr/trocr_russian_18th_century_printed
trocr_russian_18th_century_printed is a machine learning model from dsmchr. 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 cc-by-nc-4.0.
This model is a fine-tuned version of microsoft/trocr-base-printed for optical character recognition (OCR) of historical Russian printed texts from the eighteenth century.
All-time downloads
725
Public
Parameters
334M
1.3 GB on disk
Likes
0
Public
Repo makeup
Click a slice to open those files.
.safetensors1.3 GB · 100%
At a glance
- License
- cc-by-nc-4.0
- Model type
- vision-encoder-decoder
- Access
- Public
- Created
- Jul 10, 2026
- Updated
- Jul 10, 2026
- SHA
- 70e5b8a4
Model card
From the Hugging Face model README
TrOCR Base for 18th-Century Russian Printed Texts
This model is a fine-tuned version of microsoft/trocr-base-printed for optical character recognition (OCR) of historical Russian printed texts from the eighteenth century.
Intended Use
The model is designed for recognizing printed Russian texts with prereform orthography, including historical character forms and typographic features typical of eighteenth-century publications.
Base-usage
from PIL import Image
from transformers import TrOCRProcessor, VisionEncoderDecoderModel
hf_model = VisionEncoderDecoderModel.from_pretrained("dsmchr/trocr_russian_18th_century_printed")
image = Image.open("./path/to/your/image")
processor = TrOCRProcessor.from_pretrained("microsoft/trocr-base-printed")
pixel_values = processor(images=image, return_tensors="pt").pixel_values
generated_ids = hf_model.generate(pixel_values)
generated_text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
print(generated_text)
Training Dataset
The model was fine-tuned on a custom OCR dataset containing 14,911 manually aligned text lines.
The dataset was compiled from Russian printed books of various genres and printing houses, predominantly published during the second half of the eighteenth century. Ground-truth transcriptions preserve the original historical spelling.
Training Details
| Parameter | Value |
|---|---|
| Base model | microsoft/trocr-base-printed |
| Epochs | 3 |
| Batch size | 8 |
| Maximum training steps | 4473 |
| Evaluation interval | Every 200 steps |
| Best checkpoint | Step 3600 |
| Best validation CER | 0.0534 |
- Type
- vision-encoder-decoder
- License
- cc-by-nc-4.0
- Languages
- ru
- Created
- Jul 10, 2026
- Updated
- Jul 10, 2026