Layoutlmv3 Finetuned Funsd

L

Layoutlmv3 Finetuned Funsd

Developed by nielsr

A document understanding model fine-tuned on the FUNSD dataset based on the LayoutLMv3-base model, excelling in token classification tasks for forms and documents

Text Recognition

#Document Understanding #High-precision OCR #Form Recognition

Downloads 2,420

Release Time : 5/2/2022

Model Overview

This model is specifically designed for document layout analysis and understanding, capable of identifying text regions, headings, questions, answers, and other elements in documents, suitable for structured information extraction from forms and documents

Model Features

High-precision Document Understanding

Achieves 90.26% precision and 91.3% recall on the FUNSD dataset

Multimodal Processing Capability

Simultaneously processes textual content and visual layout information

End-to-End Training

Supports end-to-end processing from raw document images to structured information

Model Capabilities

Document Layout Analysis

Form Understanding

Text Region Classification

Structured Information Extraction

Use Cases

Document Processing

Form Information Extraction

Automatically extracts structured data from scanned form documents

Accurately identifies elements such as questions and answers in forms

Contract Analysis

Identifies key clauses and sections in contract documents

Enterprise Automation

Invoice Processing

Automatically extracts supplier, amount, date, and other information from invoices

🚀 layoutlmv3-finetuned-funsd

This model is a fine - tuned version of [microsoft/layoutlmv3 - base](https://huggingface.co/microsoft/layoutlmv3 - base) on the nielsr/funsd - layoutlmv3 dataset, achieving good performance in token classification tasks.

🚀 Quick Start

This model is a fine - tuned version of microsoft/layoutlmv3-base on the nielsr/funsd-layoutlmv3 dataset. It achieves the following results on the evaluation set:

Loss: 1.1164
Precision: 0.9026
Recall: 0.913
F1: 0.9078
Accuracy: 0.8330

The script for training can be found here: https://github.com/huggingface/transformers/tree/main/examples/research_projects/layoutlmv3

📚 Documentation

Tags

generated_from_trainer

Datasets

nielsr/funsd-layoutlmv3

Metrics

precision
recall
f1
accuracy

Model Index

Property	Details
Model Name	layoutlmv3-finetuned-funsd
Base Model	microsoft/layoutlmv3-base
Task Type	token - classification
Dataset Name	nielsr/funsd-layoutlmv3
Precision	0.9026198714780029
Recall	0.913
F1	0.9077802634849614
Accuracy	0.8330271015158475

Training Procedure

Training Hyperparameters

The following hyperparameters were used during training:

learning_rate: 1e - 05
train_batch_size: 16
eval_batch_size: 16
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon = 1e - 08
lr_scheduler_type: linear
training_steps: 1000

Training Results

Training Loss	Epoch	Step	Validation Loss	Precision	Recall	F1	Accuracy
No log	10.0	100	0.5238	0.8366	0.886	0.8606	0.8410
No log	20.0	200	0.6930	0.8751	0.8965	0.8857	0.8322
No log	30.0	300	0.7784	0.8902	0.908	0.8990	0.8414
No log	40.0	400	0.9056	0.8916	0.905	0.8983	0.8364
0.2429	50.0	500	1.0016	0.8954	0.9075	0.9014	0.8298
0.2429	60.0	600	1.0097	0.8899	0.897	0.8934	0.8294
0.2429	70.0	700	1.0722	0.9035	0.9085	0.9060	0.8315
0.2429	80.0	800	1.0884	0.8905	0.9105	0.9004	0.8269
0.2429	90.0	900	1.1292	0.8938	0.909	0.9013	0.8279
0.0098	100.0	1000	1.1164	0.9026	0.913	0.9078	0.8330

Framework Versions

Transformers 4.19.0.dev0
Pytorch 1.11.0+cu113
Datasets 2.0.0
Tokenizers 0.11.6

Featured Recommended AI Models

AIbase

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご

© 2025AIbase