Note that this model requires Tesseract, French package, in order to perform inference. You can install it using !sudo apt-get install tesseract-ocr-fra.

Here's how to use this model:

from transformers import AutoProcessor, AutoModelForTokenClassification
import torch
from PIL import Image

processor = AutoProcessor.from_pretrained("nielsr/layoutxlm-finetuned-xfund-fr")
model = AutoModelForTokenClassification.from_pretrained(nielsr/layoutxlm-finetuned-xfund-fr")

# assuming you have a French document, turned into an image
image = Image.open("...").convert("RGB")

# prepare for the model
encoding = processor(image, padding="max_length", max_length=512, truncation=True, return_tensors="pt")

with torch.no_grad():
  outputs = model(**encoding)
  logits = outputs.logits
  
predictions = logits.argmax(-1)

Intended uses & limitations

This model can be used for NER on French scanned documents. It can recognize 4 categories: "question", "answer", "header" and "other".

Training and evaluation data

This checkpoint used the French portion of the multilingual XFUND dataset.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 1e-05
train_batch_size: 2
eval_batch_size: 2
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_ratio: 0.1
training_steps: 1000

Training results

Framework versions

Transformers 4.22.1
Pytorch 1.10.0+cu111
Datasets 2.4.0
Tokenizers 0.12.1

3chez/layoutxlm-finetuned-xfund-fr

layoutxlm-finetuned-xfund-fr

Model usage