NafisAhmed/ShilpoBangla

---
license: cc-by-4.0
task_categories:
- image-to-text
language:
- bn
size_categories:
- 1K<n<10K
---
---
license: cc-by-4.0
task_categories:
- image-to-text
language:
- bn
size_categories:
- 1K<n<10K
pretty_name: ShilpoBangla
---

# ShilpoBangla: A Bangla Image-Text Dataset for Cultural Heritage

## Dataset Description

**ShilpoBangla** is a multimodal dataset consisting of culturally significant Bangladeshi product images paired with human-written Bangla alternative text descriptions. The dataset is designed to facilitate research in image captioning, multimodal learning, and low-resource language processing, particularly for Bangla.

The dataset captures diverse aspects of Bangladeshi cultural heritage, enabling models to learn meaningful visual-text relationships grounded in cultural context.

---

## Dataset Summary

- **Total samples:** 1,200
- **Number of categories:** 5
- **Image format:** JPG
- **Annotation format:** CSV
- **Language:** Bangla

### Categories
- Traditional Clothing
- Foods
- Musical Instruments
- Folk Arts and Crafts
- Jewelry

Each category contains **240 images**, ensuring a balanced distribution across classes.

---

## Data Fields

Each entry in the dataset is represented in the CSV annotation file with the following fields:

- **class_name** (`string`): Category label of the item
- **image_path** (`string`): Relative path formatted as `class_name/image_name.jpg`
- **alt_text_bn** (`string`): Human-written Bangla description

The descriptions include details about:
- visual appearance
- material composition
- cultural significance
- usage context

---

## Data Collection Process

The dataset was created using a combination of:

- **Crowdsourced data collection** via Google Forms
- **Direct field photography** conducted by the authors

All textual descriptions were manually written in Bangla to ensure high-quality, culturally accurate annotations.

---

## Intended Uses

This dataset can be used for:

- Bangla Image Captioning
- Image-to-Text Generation
- Multimodal Representation Learning
- Cultural AI and Digital Heritage Research

---

## Limitations

- Limited to five predefined categories
- Does not include regional dialect variations of Bangla
- Potential bias due to data collection sources

---

Citation

If you use this dataset, please cite:

@misc{s._m._nafis_ahmed_2026,
author = { S. M. Nafis Ahmed and Syed Nafees Kaiser and Muhammad Nazrul Islam },
title = { ShilpoBangla (Revision 59f6ad3) },
year = 2026,
url = { https://huggingface.co/datasets/NafisAhmed/ShilpoBangla },
doi = { 10.57967/hf/8064 },
publisher = { Hugging Face }
}
## Acknowledgements

We thank all contributors who provided images and descriptions, and acknowledge the effort involved in preserving and digitizing Bangladeshi cultural heritage.