Multilingual-Reader

in this repository we will share our works related to multilingual document reading (english,bangla and arabic).this is a work in progress,we will gradually update the repo inshaa allah

Environment Setup

DEV LOCAL ENVIRONMENT

OS          : Ubuntu 20.04.3 LTS       
Memory      : 23.4 GiB 
Processor   : Intel® Core™ i5-8250U CPU @ 1.60GHz × 8    
Graphics    : Intel® UHD Graphics 620 (Kabylake GT2)  
Gnome       : 3.36.8

python requirements

• dev - cpu - test -install

stable test environment

• Manual Setup

conda create -n mlreader python=3.8  -y
conda activate mlreader
conda install -n mlreader ipykernel --update-deps --force-reinstall -y
./install.sh

Stack

• Line based detector model: paddleOCR en-dbnet
• Word based detector model: paddleOCR ml-dbnet
• English recognizer: paddleocr - en -SVTR_LCnet
• Arabic recognizer: paddleocr - ar
• Bangla recognizer: easyocr - bn
• Word classifier : Custom
  • Kaggle Dataset Link
  • onnx conversion kernel
  • training kernel

Change-log (Dev branch)

02-06-22

• merging solved
• lang model auto download
• classifier addition
• solved negative stride issue

Docs

• docs/dev.md: dev branch doc
• weights/weights.md: custom weights integration doc

Demo

• see : demo.ipynb