Dataset · visheratin
LAION-COCO image captions in 200 languages
visheratin/laion-coco-nllb
A multilingual corpus of image captions translated from English into 200 languages by the NLLB-200 model.
What's inside
The dataset includes between 100 thousand and 1 million records (parquet files in the repository) based on the LAION-COCO sample. Each record contains a unique identifier, a link to the source image, the original English caption, and a list of translations into 200 languages from the Flores-200 set, performed by the large 3.3-billion-parameter NLLB-200 model.
What you can do with it
- Training multilingual image captioning (image-to-text) models
- Training multimodal embeddings and machine translation systems for short captions
- Tasks
- Image → text, Translation
- Languages
- Acehnese, acm, acq, Tunisian Arabic, Afrikaans, apc, Akan, Albanian and 4 more
- Size
- 100K–1M rows
- Downloads this month
- 1.7K
- Likes
- 46
Models trained on it
- visheratin/nllb-clip-large-siglipZero shot image classification 1K
- visheratin/nllb-siglip-mrl-largeZero shot image classification 565
- visheratin/nllb-clip-base-siglipZero shot image classification 482
- visheratin/nllb-siglip-mrl-baseZero shot image classification 228
- visheratin/nllb-clip-base-ocZero shot image classification 178
- oddadmix/Nawah-VL-25MImage → text 31
- visheratin/nllb-clip-base 28
- oddadmix/Nawah-VL-50MImage → text 24
How to load it
from datasets import load_dataset
ds = load_dataset("visheratin/laion-coco-nllb")