CORD: A Consolidated Receipt Dataset for Post-OCR Parsing

CORD dataset is cloned from clovaai GitHub repo

Box coordinates are normalized against image width/height
Labels with very few occurrences are replaced with O:

replacing_labels = ['menu.etc', 'menu.itemsubtotal', 
                    'menu.sub_etc', 'menu.sub_unitprice', 
                    'menu.vatyn', 'void_menu.nm', 
                    'void_menu.price', 'sub_total.othersvc_price']

Check for more info Sparrow

Citation

CORD: A Consolidated Receipt Dataset for Post-OCR Parsing

@article{park2019cord,
  title={CORD: A Consolidated Receipt Dataset for Post-OCR Parsing},
  author={Park, Seunghyun and Shin, Seung and Lee, Bado and Lee, Junyeop and Surh, Jaeheung and Seo, Minjoon and Lee, Hwalsuk}
  booktitle={Document Intelligence Workshop at Neural Information Processing Systems}
  year={2019}
}

Post-OCR parsing: building simple and robust parser via BIO tagging

@article{hwang2019post,
  title={Post-OCR parsing: building simple and robust parser via BIO tagging},
  author={Hwang, Wonseok and Kim, Seonghyeon and Yim, Jinyeong and Seo, Minjoon and Park, Seunghyun and Park, Sungrae and Lee, Junyeop and Lee, Bado and Lee, Hwalsuk}
  booktitle={Document Intelligence Workshop at Neural Information Processing Systems}
  year={2019}
}