Skip to content

Code for ICCV 2023 Paper : “ICL-D3IE: In-Context Learning with Diverse Demonstrations Updating for Document Information Extraction”

Notifications You must be signed in to change notification settings

MAEHCM/ICL-D3IE

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

19 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

ICL-D3IE: In-Context Learning with Diverse Demonstrations Updating for Document Information Extraction

Congratulations on our paper being accepted by ICCV 2023! We will continue to update the repository.

Table of contents

We propose a simple but effective in-context learning framework called ICL-D3IE, which enables LLMs to perform DIE with different types of demonstration examples.

Installation

Installation for Project,if you need to study the robustness of the model to text shift, you need to install Textattack

git clone https://anonymous.4open.science/r/ICL-D3IE-B1EE && cd ICL-D3IE

Datasets

Dataset Link
FUNSD download
CORD download
SROIE download

Preprocess Input

The data is processed as follows, like "****.json",set your input path and run preprocess_{ }.py.

{text:"TAX 5.455",Box:[490 743 819 777]}{text:"TOTAL 60.000",Box:[101 820 851 858]}{text:"(Qty 2.00",Box:[314 820 615 856]}{text:"EDC CIMB NIAGA No: xx7730 60.000",Box:[138 898 847 938]}{text:"901016",Box:[97 604 212 635]}...

Preprocess GT

The data is processed as follows, like "****.json",set your GT path and run preprocess_{ }.py.

{text:"TAX 5.455",Box:[490 743 819 777],entity:SUB_TOTAL.TAX_PRICE}{text:"TOTAL 60.000",Box:[101 820 851 858],entity:TOTAL.TOTAL_PRICE}{text:"(Qty 2.00",Box:[314 820 615 856],entity:TOTAL.MENUQTY_CNT}{text:"EDC CIMB NIAGA No: xx7730 60.000",Box:[138 898 847 938],entity:TOTAL.CREDITCARDPRICE}{text:"901016",Box:[97 604 212 635],entity:MENU.NUM}

Predict

Use GPT3 to predict the ouput perfectly!

cd GPT3 && python gpt3_{ }.py 

Or use ChatGPT to predict the ouput perfectly!

cd chatgpt && python chatgpt_{ }.py

Eval

cd eval && python eval_{ }.py

Results

Reulst of comparing ICL-D3IE with Standard ICL and existing pre-trained VDU models fine-tuned with full training samples and a few sample on three benchmark datasets in ID and OOD settings.

About

Code for ICCV 2023 Paper : “ICL-D3IE: In-Context Learning with Diverse Demonstrations Updating for Document Information Extraction”

Topics

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages