Key information extraction from invoice document with Graph Convolution Network

• Related blog post from my Viblo account: https://viblo.asia/p/djeZ1yPGZWz

• Background subtraction: U2Net
• Image alignment: computer vision techniques, cv2
• Text detection: CRAFT and an in-house text-detection model
• Text recognition: VietOCR and an in-house text-recognition model
• KIE: Graph Convolution

Currently, I dont have the invoice-direction classifier model. But you can also develop a model to rotate the image if the image is rotated horizontally or upside down.

• Google Drive

• MC-OCR, a Vietnamese receipts dataset: https://aihub.vn/competitions/1
• Preprocessed data: Google Drive

TODO

• Create virtual environment using conda or virtualenv

# with virtualenv
virtualenv -p python3 invoice_env
# activate environment
source invoice_env/bin/activate
# install prerequisite libraries
pip install -r requirements.txt

# 1st command, run API
make serve
# 2nd command, run web-gui with streamlit
make runapp

Then access the localhost server at: 0.0.0.0:7778

• Add preprocess data script

• MC-OCR dataset: https://aihub.vn/competitions/1
• U2Net: https://github.com/xuebinqin/U-2-Net
• CRAFT: https://github.com/clovaai/CRAFT-pytorch
• VietOCR: https://github.com/pbcquoc/vietocr
• Benchmarking GNNs: https://github.com/graphdeeplearning/benchmarking-gnns
• PaddleOCR: https://github.com/PaddlePaddle/PaddleOCR