# ICDAR 2025 Competition on End-to-end Document Image Machine Translation (OCR-free Track)

This is the official baseline code repository for [***ICDAR 2025 Competition on End-to-end Document Image Machine Translation (OCR-free Track)***](https://cip-documentai.github.io/)

## Dataset Download
The dataset can be downloaded from this [huggingface link](https://huggingface.co/datasets/liangyupu/DoTA_dataset).

## Baseline Implementation
This is an implementation of a simple end-to-end document image machine translation model with an image encoder and a translation decoder.
Details can be found in [***Document Image Machine Translation with Dynamic Multi-pre-trained Models Assembling (NAACL 2024 Main)***](https://aclanthology.org/2024.naacl-long.392/) Section 5.3 Base.

### 1. Requirements
```bash
python==3.10.13
pytorch==1.13.1
transformers==4.33.2
```

### 2. Download Pre-trained models
Download pre-trained Donut model from [naver-clova-ix/donut-base](https://huggingface.co/naver-clova-ix/donut-base).

Download pre-trained Nougat model from [facebook/nougat-small](https://huggingface.co/facebook/nougat-small).

### 3. Train
```bash
bash launch_train.sh
```

### 5. Inference
Before running the script, you need to replace the `~/anaconda3/envs/your_env_name/lib/python3.10/site-packages/transformers/models/bert/modeling_bert.py` file with the `modeling_bert.py` file.
```bash
bash launch_inference.sh
```

## Citation
If you want to use our dataset and cite our paper, please use the following BibTex entries:
```BibTex
@inproceedings{liang2024document,
  title={Document Image Machine Translation with Dynamic Multi-pre-trained Models Assembling},
  author={Liang, Yupu and Zhang, Yaping and Ma, Cong and Zhang, Zhiyang and Zhao, Yang and Xiang, Lu and Zong, Chengqing and Zhou, Yu},
  booktitle={Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies},
  pages={xxxx--xxxx},
  year={2024}
}
```

If you have any question, feel free to contact [dimt2025.contact@gmail.com](mailto:dimt2025.contact@gmail.com).