metadata

language:
  - vi
  - en
pretty_name: COCO 2017 image captions in Vietnamese
source-datasets:
  - ms coco
tags:
  - coco
  - coco-2017-vi
  - Vietnamese
license: unknown
task_categories:
  - image-to-text
task_ids:
  - image-captioning

COCO 2017 image captions in Vietnamese

The dataset is firstly introduced in dinhanhx/VisualRoBERTa. I use VinAI tools to translate COCO 2027 image caption (2017 Train/Val annotations) from English to Vietnamese. Then we merge UIT-ViIC dataset into it. To load the dataset, one can take a look at this code in VisualRoBERTa or this code in Velvet.

I provide both English original and Vietnamese version (including UIT-ViIC).

⚠ Note:

UIT-ViIC splits are originated from en/captions_train2017.json. Therefore, I combine all UIT-ViIC splits then I merge into vi/captions_train2017_trans.json. As a result, I get captions_train2017_trans_plus.json.
vi/captions_train2017_trans.json and vi/captions_val2017_trans.json are VinAI-translated from the ones in en/.