EditTrans

Edit-based Speedy Detailed Transformation of Academic Documents into Markup

Repository Structure

root
├── data
│   ├── rainbow_bank           >>  sample data and arXiv ids to build the complete dataset
│   ├── data_maker.py          >>  make dataset from output of LaTeX Rainbow
│   └── data_spliter.py        >>  split dataset
├── filter_pth                 >>  fine-tuned ERNIE Layout weight
├── training_ernie.py          >>  fine-tuning filter
├── nougat                     >>  implementation of metrics from Nougat model
├── edit_trains                >>  implementation of EditTrans
├── results                    >>  latency, step and scores in paper
└── test_*.py                  >>  get scores for the baseline and EditTrans

Examples

Please see data/rainbow_bank/images/ for sample images. You can find the example of data annotation here.

Acknowledgments

Dataset is annotated by LaTeX Rainbow.

This repository builds on top of the Nougat and Token-Path-Prediction repository.