Layoutlmv2 Finetuned Sroie

Developed by Theivaprakasham

A document information extraction model fine-tuned on the SROIE dataset based on the LayoutLMv2 architecture, excelling at extracting key fields from receipt documents

Sequence Labeling

Transformers

#Document Information Extraction #High-precision OCR #Invoice Recognition

Downloads 71

Release Time : 3/2/2022

Model Overview

This model is specifically designed to extract structured information from scanned receipts or invoices, such as company name, address, date, and total amount

Model Features

High-precision Field Extraction

Achieves over 95% F1 score in key receipt field extraction tasks

Multimodal Understanding

Combines text content and document layout information for comprehensive analysis

Domain Adaptation

Specifically optimized for receipt information extraction scenarios

Model Capabilities

Receipt text recognition

Structured information extraction

Document layout analysis

Multi-field joint recognition

Use Cases

Financial Automation

Receipt Information Digitization

Automatically extracts key business information from scanned receipts

Reduces manual data entry workload with over 95% accuracy

Document Processing

Invoice Information Extraction

Batch processes invoice documents and extracts structured data

Supports efficient extraction of key fields like company, date, and amount

license: cc-by-nc-sa-4.0 tags:

generated_from_trainer datasets:
sroie model-index:
name: layoutlmv2-finetuned-sroie results: []

layoutlmv2-finetuned-sroie

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

Loss: 0.0291
Address Precision: 0.9341
Address Recall: 0.9395
Address F1: 0.9368
Address Number: 347
Company Precision: 0.9570
Company Recall: 0.9625
Company F1: 0.9598
Company Number: 347
Date Precision: 0.9885
Date Recall: 0.9885
Date F1: 0.9885
Date Number: 347
Total Precision: 0.9253
Total Recall: 0.9280
Total F1: 0.9266
Total Number: 347
Overall Precision: 0.9512
Overall Recall: 0.9546
Overall F1: 0.9529
Overall Accuracy: 0.9961

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 5e-05
train_batch_size: 8
eval_batch_size: 8
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: 3000
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Address Precision	Address Recall	Address F1	Address Number	Company Precision	Company Recall	Company F1	Company Number	Date Precision	Date Recall	Date F1	Date Number	Total Precision	Total Recall	Total F1	Total Number	Overall Precision	Overall Recall	Overall F1	Overall Accuracy
No log	0.05	157	0.8162	0.3670	0.7233	0.4869	347	0.0617	0.0144	0.0234	347	0.0	0.0	0.0	347	0.0	0.0	0.0	347	0.3346	0.1844	0.2378	0.9342
No log	1.05	314	0.3490	0.8564	0.8934	0.8745	347	0.8610	0.9280	0.8932	347	0.7297	0.8559	0.7878	347	0.0	0.0	0.0	347	0.8128	0.6693	0.7341	0.9826
No log	2.05	471	0.1845	0.7970	0.9049	0.8475	347	0.9211	0.9424	0.9316	347	0.9885	0.9885	0.9885	347	0.0	0.0	0.0	347	0.8978	0.7089	0.7923	0.9835
0.7027	3.05	628	0.1194	0.9040	0.9222	0.9130	347	0.8880	0.9135	0.9006	347	0.9885	0.9885	0.9885	347	0.0	0.0	0.0	347	0.9263	0.7061	0.8013	0.9853
0.7027	4.05	785	0.0762	0.9397	0.9424	0.9410	347	0.8889	0.9222	0.9052	347	0.9885	0.9885	0.9885	347	0.7740	0.9078	0.8355	347	0.8926	0.9402	0.9158	0.9928
0.7027	5.05	942	0.0564	0.9282	0.9308	0.9295	347	0.9296	0.9510	0.9402	347	0.9885	0.9885	0.9885	347	0.7801	0.8588	0.8176	347	0.9036	0.9323	0.9177	0.9946
0.0935	6.05	1099	0.0548	0.9222	0.9222	0.9222	347	0.6975	0.7378	0.7171	347	0.9885	0.9885	0.9885	347	0.8608	0.8732	0.8670	347	0.8648	0.8804	0.8725	0.9921
0.0935	7.05	1256	0.0410	0.92	0.9280	0.9240	347	0.9486	0.9568	0.9527	347	0.9885	0.9885	0.9885	347	0.9091	0.9222	0.9156	347	0.9414	0.9488	0.9451	0.9961
0.0935	8.05	1413	0.0369	0.9368	0.9395	0.9381	347	0.9569	0.9597	0.9583	347	0.9772	0.9885	0.9828	347	0.9143	0.9222	0.9182	347	0.9463	0.9524	0.9494	0.9960
0.038	9.05	1570	0.0343	0.9282	0.9308	0.9295	347	0.9624	0.9597	0.9610	347	0.9885	0.9885	0.9885	347	0.9206	0.9020	0.9112	347	0.9500	0.9452	0.9476	0.9958
0.038	10.05	1727	0.0317	0.9395	0.9395	0.9395	347	0.9598	0.9625	0.9612	347	0.9885	0.9885	0.9885	347	0.9280	0.9280	0.9280	347	0.9539	0.9546	0.9543	0.9963
0.038	11.05	1884	0.0312	0.9368	0.9395	0.9381	347	0.9514	0.9597	0.9555	347	0.9885	0.9885	0.9885	347	0.9226	0.9280	0.9253	347	0.9498	0.9539	0.9518	0.9960
0.0236	12.05	2041	0.0318	0.9368	0.9395	0.9381	347	0.9570	0.9625	0.9598	347	0.9885	0.9885	0.9885	347	0.9043	0.8991	0.9017	347	0.9467	0.9474	0.9471	0.9956
0.0236	13.05	2198	0.0291	0.9337	0.9337	0.9337	347	0.9598	0.9625	0.9612	347	0.9885	0.9885	0.9885	347	0.9164	0.9164	0.9164	347	0.9496	0.9503	0.9499	0.9960
0.0236	14.05	2355	0.0300	0.9286	0.9366	0.9326	347	0.9459	0.9568	0.9513	347	0.9885	0.9885	0.9885	347	0.9275	0.9222	0.9249	347	0.9476	0.9510	0.9493	0.9959
0.0178	15.05	2512	0.0307	0.9366	0.9366	0.9366	347	0.9513	0.9568	0.9540	347	0.9885	0.9885	0.9885	347	0.9275	0.9222	0.9249	347	0.9510	0.9510	0.9510	0.9959
0.0178	16.05	2669	0.0300	0.9312	0.9366	0.9339	347	0.9543	0.9625	0.9584	347	0.9885	0.9885	0.9885	347	0.9171	0.9251	0.9211	347	0.9477	0.9532	0.9504	0.9959
0.0178	17.05	2826	0.0292	0.9368	0.9395	0.9381	347	0.9570	0.9625	0.9598	347	0.9885	0.9885	0.9885	347	0.9253	0.9280	0.9266	347	0.9519	0.9546	0.9532	0.9961
0.0178	18.05	2983	0.0291	0.9341	0.9395	0.9368	347	0.9570	0.9625	0.9598	347	0.9885	0.9885	0.9885	347	0.9253	0.9280	0.9266	347	0.9512	0.9546	0.9529	0.9961
0.0149	19.01	3000	0.0291	0.9341	0.9395	0.9368	347	0.9570	0.9625	0.9598	347	0.9885	0.9885	0.9885	347	0.9253	0.9280	0.9266	347	0.9512	0.9546	0.9529	0.9961

Framework versions

Transformers 4.16.2
Pytorch 1.8.0+cu101
Datasets 1.18.4.dev0
Tokenizers 0.11.6

Featured Recommended AI Models

Empowering the Future, Your AI Solution Knowledge Base

English 简体中文繁體中文にほんご