├── README.md
├── README_EN.md
├── com.png
├── model.png
└── sota.png
/README.md:
--------------------------------------------------------------------------------
1 |
2 |
3 |
4 |
5 |
8 |
9 | [简体中文](README.md) | [English](README_EN.md) | [Paper](https://arxiv.org/abs/2308.06743)
10 | # TextDiff: Mask-Guided Residual Diffusion Models for Scene Text Image Super-Resolution
11 | 这里是论文[TextDiff: Mask-Guided Residual Diffusion Models for Scene Text Image Super-Resolution]((https://arxiv.org/abs/2308.06743))的官方复现仓库。TextDiff是一个场景文字超分辨率优化模型（详见[论文](https://arxiv.org/abs/2308.06743)).
12 |
13 |
14 | # 网络结构
15 |
16 |
17 |
18 |
见config.yaml文件
55 |
56 | 3. 训练
57 | ```
58 | python train.py
59 | ```
60 | ## 推理
61 | ```
62 | python test.py
63 | ```
64 |
65 | # To-do lists
66 |
67 | - [ ] 添加训练代码(To be released soon.)
68 | - [ ] 添加推理代码(To be released soon.)
69 | - [ ] 使用DPM_solver减少推理步长
70 |
71 |
72 | # 效果图
73 |
74 |
75 | # 感谢
76 |
77 | - 如果你觉得TextDiff对你有帮助，请给个star，谢谢！
78 | - 如果你有任何问题，欢迎提issue(issue通知与我邮箱绑定，看到后我会及时回复)。
79 | - 如果你愿意将TextDiff作为你的项目的baseline，欢迎引用我们的论文。
80 |
81 |
82 | # References
83 |
84 |
85 | - [1] Scene text telescope:
86 | Text-focused scene image super-resolution
87 | - [2] Activating more pixels in image super-resolution
88 | transformer.
89 | - [3] Srdiff: Single image super-resolution
90 | with diffusion probabilistic models.
91 | - [4] DocDiff: Document Enhancement via Residual Diffusion Models
92 | - [5] Improving
93 | Scene Text Image Super-Resolution via Dual Prior Modulation Network
94 |
95 |
96 | # :book: Citation
97 | If you use (part of) my code or find my work helpful, please consider citing
98 | ```
99 | @article{liu2023textdiff,
100 | title={TextDiff: Mask-Guided Residual Diffusion Models for Scene Text Image Super-Resolution},
101 | author={Liu, Baolin and Yang, Zongyuan and Wang, Pengfei and Zhou, Junjie and Liu, Ziqi and Song, Ziyi and Liu, Yan and Xiong, Yongping},
102 | journal={arXiv preprint arXiv:2308.06743},
103 | year={2023}
104 | }
105 | ```
106 | # Acknowledgement
107 | This code is developed relying on DocDiff and TATT. Thanks for these great projects. Among them, DocDiff is the main research content of my classmate, and I participated in part of the research.
108 |
--------------------------------------------------------------------------------
/README_EN.md:
--------------------------------------------------------------------------------
1 |
2 |
3 |
4 |
5 |
8 |
9 | [简体中文](README.md) | [English](README_EN.md) | [Paper](https://arxiv.org/abs/2308.06743)
10 | # TextDiff: Mask-Guided Residual Diffusion Models for Scene Text Image Super-Resolution
11 | Here is the official reproduction repository of the paper [TextDiff: Mask-Guided Residual Diffusion Models for Scene Text Image Super-Resolution]((https://arxiv.org/abs/2308.06743)). TextDiff is a scene text super-resolution optimization model (see [paper](https://arxiv.org/abs/2308.06743) for details).
12 |
13 |
14 | # Network Structure
15 |
16 |
17 |
18 |
