Hugging Face's logo

Datasets:
MBZUAI
/
DuwatBench

Modalities:
Image
Text
Formats:
parquet
Size:
1K - 10K
ArXiv:
Libraries:
Datasets
pandas
Polars
License:
Dataset card Data Studio Files
xet
Community
1
DuwatBench / README.md
SLMLAH's picture
SLMLAH
Update README.md
11d15cb verified
preview code
|
raw
history blame contribute delete
7.94 kB
 ---
configs:
- config_name: default
data_files:
- split: train
path: data/train-*
license: apache-2.0
dataset_info:
features:
- name: image
dtype: image
- name: image_id
dtype: string
- name: style
dtype: string
- name: category
dtype: string
- name: source
dtype: string
- name: text
list: string
- name: word_count
list: int32
- name: total_words
dtype: int32
- name: bboxes
list:
list: int32
splits:
- name: train
num_bytes: 402789351
num_examples: 1272
download_size: 399125040
dataset_size: 402789351
---
<div align="center">
<img src='figures/logo.png' align="left" width="12%" />
</div>
<div style="margin-top:50px;">
<h1 style="font-size: 30px; margin: 0;"> DuwatBench: Bridging Language and Visual Heritage through an Arabic Calligraphy Benchmark for Multimodal Understanding [EACL 2026 (Main)🔥]</h1>
</div>
<div align="center" style="margin-top:10px;">
[Shubham Patle](https://github.com/shubhamrpatle) <sup>1* </sup> &nbsp;
[Sara Ghaboura](https://huggingface.co/SLMLAH) <sup>1* </sup> &nbsp;
[Hania Tariq](https://huggingface.co/) <sup>2 </sup> &nbsp;
[Mohammad Usman Khan](https://huggingface.co/) <sup>3 </sup> &nbsp;
<br>
[Omkar Thawakar](https://omkarthawakar.github.io) <sup>1 </sup>&nbsp;
[Rao M. Anwer](https://scholar.google.com/citations?hl=en&user=_KlvMVoAAAAJ) <sup>1</sup>&nbsp;
[Salman Khan](https://scholar.google.com/citations?hl=en&user=M59O9lkAAAAJ) <sup>1,4 </sup>
<br>
<br>
<p align="center">
<sup>1</sup>Mohamed bin Zayed University of AI &nbsp;&nbsp;
<sup>2</sup>NUCES &nbsp;&nbsp;
<sup>3</sup>NUST &nbsp;&nbsp;
<sup>4</sup>Australian National University
</p>
[![arXiv](https://img.shields.io/badge/arXiv-2601.19898-FA7758)](https://arxiv.org/abs/2601.19898)
[![Our Page](https://img.shields.io/badge/Visit-Our%20Page-FA9B58?style=flat)](https://mbzuai-oryx.github.io/DuwatBench/)
[![Code](https://img.shields.io/badge/code-GitHub-FAD358?style=flat)](https://github.com/mbzuai-oryx/DuwatBench)
[![GitHub license](https://img.shields.io/github/license/mbzuai-oryx/DuwatBench?color=CDCDCD)](https://github.com/mbzuai-oryx/DuwatBench/blob/main/LICENSE)
<br>
<em> <sup> *Equal Contribution </sup> </em>
</div>
<p align="center">
<img src="figures/line.png" height="9px">
</p>
<div align="center">
<b> If you like our project, please like it ❤️ and give us a star ⭐ on GitHub. </b><br>
</div>
<p align="center">
<img src="figures/line.png" height="9px">
</p>
## Latest Updates
🔥🔥 **[04 Jan 2026]** 🔥🔥 DuwatBench accepted to EACL 2026 Main track.<br>
🔥 **[22 Jan 2026]** DuwatBench, the Arabic Calligraphy Benchmark for Multimodal Understanding is released.<br>
🤗 **[23 Jan 2026]** DuwatBench dataset available on [HuggingFace](https://huggingface.co/datasets/MBZUAI/TimeTravel).<br>
---
## ✨ Overview
<div align="center">
**DuwatBench** is a comprehensive benchmark for evaluating LMMs on Arabic calligraphy recognition. Arabic calligraphy represents one of the richest visual traditions of the Arabic language, blending linguistic meaning with artistic form. DuwatBench addresses the gap in evaluating how well modern AI systems can process stylized Arabic text.
<br>
| <h6><em> Fig.1.a. Proportional breakdown of calligraphic styles<br>in the DuwatBench dataset </h6></em> | <h6><em> Fig.1.b.Proportional breakdown of textual categories,<br>covering religious and non-religious themes </h6></em> |
|:-------------------------------------------------------------------------:|:-------------------------------------------------------------------------------------------:|
| <img src="figures/style_stat.png" width="300"/> | <img src="figures/cat_stat.png" width="300"/> |
</div>
</p>
---
## 🌟 Key Features
### **Key Features of TimeTravel**
- **1,272 curated samples** spanning 6 classical and modern calligraphic styles
- **Over 9.5k word instances with approximately 1,475 unique words** spanning religious and cultural domains
- **Bounding box annotations** for detection-level evaluation
- **Full text transcriptions** with style and theme labels
- **Complex artistic backgrounds** preserving real-world visual complexity
---
## 🏗️ DuwatBench Creation Pipeline
The DuwatBench dataset follows a structured pipeline to ensure the accuracy, completeness, and contextual richness by style and category.<br>
<p align="center">
<img src="figures/pipeline.png" width="2300px" height="250px" alt="pipeline" style="margin-right: 2px";/>
<h6>
<em> <b>Figure 2. End-to-end pipeline for constructing DuwatBench, from data collection and manual transcription with bounding boxes to multi-tier verification and style/theme aggregation.</b> </em>
</h6>
---
## ✒️ Calligraphic Styles
| Style | Arabic | Description |
|-------|--------|-------------|
| **Thuluth** | الثلث | Ornate script used in mosque decorations |
| **Diwani** | الديواني | Flowing Ottoman court script |
| **Naskh** | النسخ | Standard readable script |
| **Kufic** | الكوفي | Geometric angular early Arabic script |
| **Ruq'ah** | الرقعة| Modern everyday handwriting |
| **Nasta'liq** | النستعليق| Persian-influenced flowing script |
---
## 🧐 DuwatBench Dataset Examples
<p align="center">
<img src="figures/samples.png" width="2300px" height="400px" alt="sample" style="margin-right: 2px";/>
<h6>
<em> <b>Figure 2. End-to-end pipeline for constructing DuwatBench, from data collection and manual transcription with bounding boxes to multi-tier verification and style/theme aggregation.</b> </em>
</h6>
---
## 🔍 Qualitative Evaluation and Results
<p align="center">
<img src="figures/qualitative.png" width="2300px" height="700px" alt="pipeline" style="margin-right: 2px";/>
<h6>
<em> <b>Figure 3. Qualitative results comparing open- and closed-source models on DuwatBench calligraphy samples.</b> </em>
</h6>
---
## 📚 Citation
If you use DuwatBench dataset in your research, please consider citing:
```bibtex
@misc{patle2026duwatbenchbridginglanguagevisual,
title={DuwatBench: Bridging Language and Visual Heritage through an Arabic Calligraphy Benchmark for Multimodal Understanding},
author={Shubham Patle and Sara Ghaboura and Hania Tariq and Mohammad Usman Khan and Omkar Thawakar and Rao Muhammad Anwer and Salman Khan},
year={2026},
eprint={2601.19898},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2601.19898},
}
```
---
## ⚖️ License
This project is licensed under the Apache License 2.0 - see the [LICENSE](LICENSE) file for details.
The dataset images are sourced from public digital archives and community repositories under their respective licenses.
---
## 🙏 Acknowledgments
- Digital archives: [Library of Congress](https://www.loc.gov/collections/), [NYPL Digital Collections](https://digitalcollections.nypl.org/)
- Community repositories: [Calligraphy Qalam](https://calligraphyqalam.com/), [Free Islamic Calligraphy](https://freeislamiccalligraphy.com/), [Pinterest](https://www.pinterest.com/)
- Annotation tool: [MakeSense.ai](https://www.makesense.ai/)
- Arabic NLP tools: [CAMeL Tools](https://github.com/CAMeL-Lab/camel_tools)
---
## 📧 Contact
For questions or issues, please:
- Open an issue on [GitHub](https://github.com/mbzuai-oryx/DuwatBench/issues)
- Contact the authors at: {shubham.patle, sara.ghaboura, omkar.thawakar}@mbzuai.ac.ae
---
<p align="center">
<img src="figures/IVAL_logo.png" width="18%" style="display: inline-block; margin: 0 10px;" />
<img src="figures/Oryx_logo.jpeg" width="10%" style="display: inline-block; margin: 0 10px;" />
<img src="figures/MBZUAI_Logo_EN_Blue_CMYK.jpg" width="30%" style="display: inline-block; margin: 0 10px;" />
</p>