StructuralLM

StructuralLM: Structural Pre-training for Form Understanding

https://arxiv.org/abs/2105.11210

Introduction

This work presents a Structural LM model for document image understanding. We introduce a new pre-training approach to jointly leverage cell and layout information from scanned documents.

Pre-trained models

Model	Description	#params	Download
structurallm.en.large	StructuralLM using the BERT-large architecture	340M	Structural lm model

Example usage

Requirements and Installation

• Tensorflow version == 1.14.0
• Install other libraries via

pip install -r requirements.txt

Some codes are borrowed from Bert

Finetune FUNSD dataset

Download the processed data (funsd data)

sh finetune_funsd_dataset.sh 

Other Task

Comming soon

Citation

If you use our work, please cite:

@misc{li2021structurallm,
      title={StructuralLM: Structural Pre-training for Form Understanding}, 
      author={Chenliang Li and Bin Bi and Ming Yan and Wei Wang and Songfang Huang and Fei Huang and Luo Si},
      year={2021},
      eprint={2105.11210},
      archivePrefix={arXiv},
      primaryClass={cs.CL}
}