---
dataset_info:
  features:
  - name: id
    dtype: string
  - name: hq_img
    dtype: image
  - name: lq_img_lv1
    dtype: image
  - name: lq_img_lv2
    dtype: image
  - name: lq_img_lv3
    dtype: image
  - name: text
    sequence: string
  - name: bbox
    sequence:
      array2_d:
        shape:
        - 2
        - 2
        dtype: int32
  - name: poly
    sequence:
      array2_d:
        shape:
        - 16
        - 2
        dtype: int32
  splits:
  - name: test
    num_bytes: 119459110.0
    num_examples: 1000
  download_size: 118582963
  dataset_size: 119459110.0
configs:
- config_name: default
  data_files:
  - split: test
    path: data/test-*
---

# SA-Text

**Text-Aware Image Restoration with Diffusion Models** (arXiv:2506.09993)  
Large-scale training dataset for the **Text-Aware Image Restoration (TAIR)** task.

- 📄 Paper: https://arxiv.org/abs/2506.09993  
- 🌐 Project Page: https://cvlab-kaist.github.io/TAIR/  
- 💻 GitHub: https://github.com/cvlab-kaist/TAIR  
- 🛠 Dataset Pipeline: https://github.com/paulcho98/text_restoration_dataset

## Dataset Description

The test set is organized into three degradation levels (lv1–lv3) with overlapping severity ranges, and stochastic degradation kernels make the ordering non-strict.

## Notes

- Each image includes one or more **text instances** with transcriptions and polygon-level labels.
- Designed for training **TeReDiff**, a multi-task diffusion model introduced in our paper.
- For the training set of SA-Text, check [SA-Text](https://huggingface.co/datasets/Min-Jaewon/SA-Text)
- For real-world evaluation, check [Real-Text](https://huggingface.co/datasets/Min-Jaewon/Real-Text).
  
## Citation
Please cite the following paper if you use this dataset:
```bibtex
@article{min2024textaware,
  title={Text-Aware Image Restoration with Diffusion Models},
  author={Min, Jaewon and Kim, Jin Hyeon and Cho, Paul Hyunbin and Lee, Jaeeun and Park, Jihye and Park, Minkyu and Kim, Sangpil and Park, Hyunhee and Kim, Seungryong},
  journal={arXiv preprint arXiv:2506.09993},
  year={2025}
}