yeppeuda13/TrOCR_Herbiome

TrOCR Herbiome — Herbarium Label OCR

A fine-tuned version of agomberto/trocr-large-handwritten-fr for transcription of text found on European and French herbarium specimen labels.

Model Description

This model was fine-tuned to handle the specific OCR challenges of herbarium digitization. Herbarium labels are particularly difficult because a single label often contains both printed institutional text and handwritten collector notes, sometimes dating back to the 19th or early 20th century.

The model handles:

• Handwritten French and Latin botanical text
• Printed institutional labels
• Mixed handwritten and printed content on the same label
• Historical script styles common in European herbarium collections

Training Data

The model was fine-tuned on a private dataset of annotated French and European herbarium label images. The dataset is not publicly available.

Intended Use

This model is designed to be used as part of the HERBIOME pipeline, which combines YOLO-based component detection and CRAFT-based text region detection before passing word crops to this model for transcription.

It is specifically optimized for:

• European herbarium collections
• French-language labels
• Word-level recognition (crops of individual words, not full labels)

Evaluation

The model is evaluated using:

• CER (Character Error Rate)
• WER (Word Error Rate)

Benchmarks were computed on a held-out subset of the IRD/UMMISCO herbarium label dataset.

How to Use

from transformers import TrOCRProcessor, VisionEncoderDecoderModel
from PIL import Image

processor = TrOCRProcessor.from_pretrained("yeppeuda13/TrOCR_Herbiome")
model = VisionEncoderDecoderModel.from_pretrained("yeppeuda13/TrOCR_Herbiome")

image = Image.open("word_crop.jpg").convert("RGB")
pixel_values = processor(images=image, return_tensors="pt").pixel_values

generated_ids = model.generate(pixel_values)
text = processor.batch_decode(generated_ids, skip_special_tokens=True)[0]
print(text)

Acknowledgements

This work was carried out as part of the HERBIOME project.

Citation

If you use this model, please credit the HERBIOME project.