layoutlmv2-document-classifier

This model is a fine-tuned version of microsoft/layoutlmv2-base-uncased on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.0008
• Accuracy: 1.0

Dataset Infomation

This model was fine-tuned to classify some company documents.

Dataset used: Company Documents Dataset

Dependencies

pip install PyMuPDF
pip install transformers
pip install torch
pip install torchvision
pip install pytesseract

• setup tesseract locally in your machine follow steps here: install instructions

Model Usage

use a file in this dataset to test: https://www.kaggle.com/datasets/navodpeiris/company-documents-dataset

import os
from PIL import Image
from transformers import LayoutLMv2Processor, LayoutLMv2ForSequenceClassification
import fitz
import io

processor = LayoutLMv2Processor.from_pretrained("microsoft/layoutlmv2-base-uncased")
model = LayoutLMv2ForSequenceClassification.from_pretrained("navodPeiris/layoutlmv2-document-classifier")

DATA_FOLDER = "data"
filename = "invoice.pdf"

file_location = os.path.join(DATA_FOLDER, filename)
doc = fitz.open(file_location)

page = doc.load_page(0)
pix = page.get_pixmap(dpi=200)

# Convert Pixmap to bytes
img_bytes = pix.tobytes("png")

# Load into PIL.Image
image = Image.open(io.BytesIO(img_bytes)).convert("RGB")
doc.close()

encoding = processor(image, return_tensors="pt", truncation=True, padding="max_length", max_length=512)

outputs = model(**encoding)
logits = outputs.logits

predicted_class_id = logits.argmax(dim=1).item()
classified_output = model.config.id2label[predicted_class_id]

print(f"Predicted class: {classified_output}")

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 1

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy
0.7722	0.0970	26	0.2249	0.9216
0.0828	0.1940	52	0.0452	0.9907
0.026	0.2910	78	0.0459	0.9907
0.0265	0.3881	104	0.0267	0.9907
0.0263	0.4851	130	0.0068	1.0
0.008	0.5821	156	0.0026	1.0
0.0023	0.6791	182	0.0014	1.0
0.0014	0.7761	208	0.0009	1.0
0.0011	0.8731	234	0.0008	1.0
0.0012	0.9701	260	0.0008	1.0

Framework versions

• Transformers 4.51.3
• Pytorch 2.6.0+cu124
• Datasets 3.6.0
• Tokenizers 0.21.1