tanziro/bangladeshi-food

Bangladeshi Food — Permissive Subset

An open image-classification dataset of Bangladeshi food. Every image is under a permissive Creative Commons license (CC0, CC-BY, CC-BY-SA), Public Domain, or GFDL. Safe for research and commercial use under the terms of CC BY-SA 4.0 (the most restrictive license present), provided you preserve attribution.

This is the public, redistributable subset of a larger 1,041-class Bangladeshi food taxonomy. Classes whose only available images had unverified licensing have been dropped from this release.

Coverage

• ~604 classes with permissive images
• ~2,273 image URLs, ~3.8 images per class average
• All licensed under CC-permissive terms

Includes major Bangladeshi food categories: fish species (Hilsa, Rohu, Pabda, Chitol…), fruits including ~30 mango varieties (Himsagar, Langra, Fazli, Khirsapat, Haribhanga…), vegetables (pointed gourd, bitter gourd, banana flower, drumstick, etc.), traditional sweets (Roshogolla, Rasmalai, Sandesh), pithas (Bhapa, Chitoi, Patishapta…), and signature dishes (Kachchi Biryani, Khichuri, Shorshe Ilish, Doi Mangsho).

Schema

Field	Type	Notes
image	Image	PIL-decodable image
label	ClassLabel	~604 snake_case classes
dish_name	string	Human-readable English
dish_name_bn	string	Bangla script (e.g. কালা ভুনা)
cuisine	ClassLabel	Always bangladeshi here
category	string	dish / fish / fruit / vegetable / sweet / pitha / etc.
region	string	e.g. Chittagong, Sylhet, Nationwide
description	string	1–2 sentence description
ingredients	list[string]	Key ingredients
cooking_method	string	e.g. slow-braised, steamed
food_type	ClassLabel	homemade / restaurant / raw / packaged / street_food
license	string	CC0 / CC-BY-* / Public Domain / GFDL
license_url	string	License URL
attribution	string	Photographer name from Wikimedia Commons
source_url	string	Original Wikimedia / Flickr page

License & attribution

Most images are CC BY-SA 4.0 (Wikimedia Commons standard). The complete distribution mix:

• CC BY-SA 4.0: ~1,044 images
• CC BY-SA 2.0/3.0: ~283 images
• Public Domain: ~126 images
• CC0: ~114 images
• CC BY 2.0/3.0/4.0: ~205 images
• GFDL: ~7 images

The dataset as a whole is released under CC BY-SA 4.0. If you use it, you must:

1. Attribute the original photographers — every row carries attribution and source_url. Aggregate citations to "Wikimedia Commons contributors" plus a link to your downstream work pointing back to this dataset card are acceptable.
2. Share alike — derivative datasets must also be released under CC BY-SA 4.0.

Usage

from datasets import load_dataset
ds = load_dataset("tanziro/bangladeshi-food")
print(ds)

# A sample
ex = ds["train"][0]
ex["image"].show()
print(ex["dish_name"], "|", ex["dish_name_bn"], "|", ex["region"])

How it was built

Images were retrieved from Wikimedia Commons via the MediaWiki API and from Openverse (which aggregates Wikimedia, Flickr CC, Smithsonian, and museum collections). The license filter accepts only cc0, pdm, by, and by-sa variants.

For raw produce / species categories (fish species, fruits, vegetables, spices, rice cultivars), if a query for the exact Bengali transliterated name returned nothing, a generic ingredient query was tried (e.g. "rohu" → "rohu carp fish"). For prepared dishes, only dish-specific matches were accepted; generic substitutions were rejected to avoid poisoning classification training.

Citation

@dataset{bangladeshi_food_permissive_2026,
  title  = {Bangladeshi Food (Permissive-Licensed Subset)},
  author = {{Tanzir} and Wikimedia Commons contributors},
  year   = {2026},
  url    = {https://huggingface.co/datasets/tanzir71/bangladeshi-food},
  note   = {Distributed under CC BY-SA 4.0; per-image attribution in row metadata.}
}

Individual photographers are credited per-row in the attribution field; please preserve those citations in any derivative work.