DeQA-Doc: Adapting DeQA-Score to Document Image Quality Assessment

Junjie Gao, Runze Liu, Yingzhe Peng, Shujian Yang, Jin Zhang, Kai Yang and Zhiyuan You

This repository is the official implementation of "DeQA-Doc: Adapting DeQA-Score to Document Image Quality Assessment".

Our DeQA-Doc wins the Championship 🏆 in the VQualA 2025 DIQA (Document Image Quality Assessment) Challenge.

mPLUG-Owl2-7B Training

Installation

git clone https://github.com/Junjie-Gao19/DeQA-Doc.git
cd DeQA-Doc/DeQA-Score
pip install -e .

If you want to train, you need to install extra dependencies:

pip install -e .[train]

Refer to the Readme of the DeQA repository

DeQA

Download pre-trained model

You can obtain the initial weight from mPLUG-Owl2

DIQA_model are different models trained separately in different dimensions

DeQA-Mix is a separate model trained with multiple dimensions mixed

Infer

sh scripts/infer.sh

When you finish infer, you need to use eval to transfer the result to the format of DIQA.

sh scripts/diqa_eval.sh

Train

if you want to train your own model

sh scripts/train.sh 
or
sh scripts/train_lora.sh

Qwen2.5-VL-7B Training

Use Llamafactory framwork to train Qwen2.5-VL-7B model

Install Llamafactory

Llamafactory

Exchange src files

You need to exchange the files in Llamafactory with the files in this repository. Their positions in Llama factory are consistent with those in this repository.

Train

llamafactory-cli train examples/train_full/qwen2.5_vl_diqa_sft.yaml

Infer

You should use the DeQA infer script to infer the result.

sh scripts/infer_qwen.sh

Acknowledgements

This work is based on DeQA-Score. Sincerely thanks for this awesome work.

Citation

If you find our work useful for your research and applications, please cite using the BibTeX:

@inproceedings{deqadoc,
  title={{DeQA-Doc}: Adapting {DeQA-Score} to Document Image Quality Assessment}, 
  author={Gao, Junjie and Liu, Runze and Peng, Yingzhe and Yang, Shujian and Zhang, Jin and Yang, Kai and You, Zhiyuan},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision Workshop},
  year={2025},
}