An OCR Toolbox for Vietnamese Documents

Configurations

Run locally

• Create conda environment, note that python version should be Python 3.9

conda create --name ocr-tool-box python=3.9
conda activate ocr-tool-box

• Install required packages

pip install -r requirements.txt --no-cache-dir

Pipeline

Main Pipeline

Process Flow Block

There are two stages (can also run in second stage only):

• The first stage is to detect and rectify document in the image, then forward through the "process flow" to find the best orientation of the document.

• The second stage is to forward the rotated image through the entire "process flow" normally to retrieve information

• Full pipeline:

python run.py

• Extra Parameters

  • --L: to specific the List of Videos you want.
  • --V: to specific the Video you want.

  python run.py --L 01 --V 001
  

  This will work for only on Video L01_V001.

  python run.py --L 01
  

  This will work for all Video with format L01_Vxxx.

References

• https://github.com/WenmuZhou/PAN.pytorch
• https://github.com/andrewdcampbell/OpenCV-Document-Scanner
• https://github.com/pbcquoc/vietocr
• https://github.com/kaylode/vietnamese-ocr-toolbox