table-extractor

Consists of various table-related inference calls for table reconstruction in documents. All the code is encapsulated in the 'tables' directory. The 'uploads' directory has sample images.

Setting Up

Install the required dependencies

pip install -r requirements.in

Download the model

Download sprint.pt from the Releases Section and place it in 'tables/model' directory.

Source Code Details

Following table calls are integrated in this repository

table-detection

Based on our trained Yolo model equipped for multilingual table detection.

python3 infer.py <page-image-path> td True

table-structure-recognition

Based on SPRINT, our script-agnostic table structure recognizer can predict OTSL sequences.

python3 infer.py <table-image-path> tsr True

full-page-reconstrcution

Uses YOLO-based table detector, SPRINT and Tesseract to generate an HOCR composed of text and tables in the inoput page image.

python3 infer.py <page-image-path> ocr True

Containerization

Building Image

cd tables
docker build -t tablecalls .

Running Container

docker run --rm --gpus all -it -v '/data/DHRUV/Document-OCR-App/document-layout-ocr/uploads/table.jpg':/docker/uploads/tables.jpg tablecalls uploads/table.jpg tsr False

User Interface

Uses streamlit to run all the required calls

streamlit run api.py