TransLayoutDocs

Translate documents, recreate layouts, convert PDF to Word, PaddleOCR, Image to Word.

Related Projects：

• Recognition-table-with-table-transformer: A system combining Table Transformer and vietOCR for accurate table structure and Vietnamese character recognition from images.

• Vietnamese_OCR_documents: Converts text from images or PDF scans into digital format, streamlining Vietnamese text processing and information management.

  
  

Table of Contents

Technical Overview

• Python > 3.6
• Recovery Layout: PPStructure (PaddleOCR)
• OCR: PaddleOCR
• Translation :
  • mbart-large-50-many-to-many-mmt(jp)
  • envit5-translation(vi)
• Spell Corection
  • oliverguhr/spelling-correction-english-base(en)
  • bmd1905/vietnamese-correction(vi)

Introduction

This diagram introduces the Layout Translation Extraction process we developed based on PaddleOCR, from image correction to information reconstruction. You can read the docs here to see differences

Visualization

• Image: RecoveryLayout

• Image: RecoveryLayout + Translate

• PDF standard: RecoveryLayout

Installization

• Clone this project:

git clone https://github.com/KaiKenju/TransLayoutDocs.git

• Initial enviromment with Miniconda (Default: python=3.10):

conda create -n <env_name> python=3.10

• Activate conda

conda activate <env_name> 
cd TransLayoutDocs

• Run the commands:

pip install -r requirements.txt

Quick Start

• Scan PDF (using OCR)

python main.py

• Standard PDF

python predict_pdf.py --image_dir=inputs\pdf\2009.09941v3.pdf --recovery=True --use_pdf2docx_api=True --output=./outputs/

• Image to Word

python img2docx.py --input=./inputs/imgs/check2.png --output=./detail_img --lang=vi --device=cpu

Usage: python img2docx.py --input=./inputs/imgs/chap4.png --output=./detail_img --lang=vi --device=cpu

  --input              Input img . Default: inputs
  --output             Output folder. Default: detail_img
  --lang               Languague to translate. Default: vi
  --device             Cpu or cuda. Default: cpu

• Note: Following the step if you translate in KR,JP,CH to see in terminal : here

Contributer

Special thanks to all the developers who have contributed to TransLayoutDocs!