# DocRes: A Generalist Model Toward Unifying Document Image Restoration Tasks [![Open in Spaces](https://huggingface.co/datasets/huggingface/badges/resolve/main/open-in-hf-spaces-sm.svg)](https://huggingface.co/spaces/qubvel-hf/documents-restoration)

This is the official implementation of our paper [DocRes: A Generalist Model Toward Unifying Document Image Restoration Tasks](https://arxiv.org/abs/2405.04408). ## News 🔥 [2025.7] Our paper ["Aesthetics is Cheap, Show me the Text: An Empirical Evaluation of State-of-the-Art Generative Models for OCR"](https://arxiv.org/abs/2507.15085), which conducts a comprehensive evaluation of SOTA generative models has been online at arXiv. 🔥 🔥 [2025.6] Beyond GPT-4o, we evaluate more SOTA generative models' image generation abilities in various document processing tasks. Check [here](https://github.com/NiceRingNode/Awesome-Image-Generators-for-OCR-Image-Generation-and-Editing)! 🔥 🎉 [2025.5] We evaluate the image generation ability of GPT-4o, including various document processing tasks. Check [here](https://github.com/NiceRingNode/Awesome-Image-Generators-for-OCR-Image-Generation-and-Editing)! 🔥 🎉 [2025.2] Our new work [LGGPT](https://github.com/NiceRingNode/LGGPT) has been accepted to IJCV 2025, an LLM that unifies versatile layout generation tasks! Welcome to follow! 🔥 A comprehensive [Recommendation for Document Image Processing](https://github.com/ZZZHANG-jx/Recommendations-Document-Image-Processing) is available. ## Inference 1. Put MBD model weights [mbd.pkl](https://1drv.ms/f/s!Ak15mSdV3Wy4iahoKckhDPVP5e2Czw?e=iClwdK) to `./data/MBD/checkpoint/` 2. Put DocRes model weights [docres.pkl](https://1drv.ms/f/s!Ak15mSdV3Wy4iahoKckhDPVP5e2Czw?e=iClwdK) to `./checkpoints/` 3. Run the following script and the results will be saved in `./restorted/`. We have provided some distorted examples in `./input/`. ```bash python inference.py --im_path ./input/for_dewarping.png --task dewarping --save_dtsprompt 1 ``` - `--im_path`: the path of input document image - `--task`: task that need to be executed, it must be one of _dewarping_, _deshadowing_, _appearance_, _deblurring_, _binarization_, or _end2end_ - `--save_dtsprompt`: whether to save the DTSPrompt ## Evaluation 1. Dataset preparation, see [dataset instruction](./data/README.md) 2. Put MBD model weights [mbd.pkl](https://1drv.ms/f/s!Ak15mSdV3Wy4iahoKckhDPVP5e2Czw?e=iClwdK) to `data/MBD/checkpoint/` 3. Put DocRes model weights [docres.pkl](https://1drv.ms/f/s!Ak15mSdV3Wy4iahoKckhDPVP5e2Czw?e=iClwdK) to `./checkpoints/` 2. Run the following script ```bash python eval.py --dataset realdae ``` - `--dataset`: dataset that need to be evaluated, it can be set as _dir300_, _kligler_, _jung_, _osr_, _docunet\_docaligner_, _realdae_, _tdd_, and _dibco18_. ## Training 1. Dataset preparation, see [dataset instruction](./data/README.md) 2. Specify the datasets_setting within `train.py` based on your dataset path and experimental setting. 3. Run the following script ```bash bash start_train.sh ``` ## Citation ``` @inproceedings{zhangdocres2024, Author = {Jiaxin Zhang, Dezhi Peng, Chongyu Liu , Peirong Zhang and Lianwen Jin}, Booktitle = {In Proceedings of the IEEE/CV Conference on Computer Vision and Pattern Recognition}, Title = {{DocRes: A Generalist Model Toward Unifying Document Image Restoration Tasks}}, Year = {2024}} ``` ## ⭐ Star Rising [![Star Rising](https://api.star-history.com/svg?repos=ZZZHANG-jx/DocRes&type=Timeline)](https://star-history.com/#ZZZHANG-jx/DocRes&Timeline)