README.md

metadata

language:
  - en
  - th
  - ru
  - ja
  - it
  - de
  - vi
  - zh
  - ar
  - es
license: cc-by-4.0
size_categories:
  - 100K<n<1M
task_categories:
  - image-to-text
pretty_name: multilingual-coco
dataset_info:
  features:
    - name: cocoid
      dtype: int64
    - name: filename
      dtype: string
    - name: en
      sequence: string
    - name: th
      sequence: string
    - name: ru
      sequence: string
    - name: jp-stair
      sequence: string
    - name: it
      sequence: string
    - name: de
      sequence: string
    - name: vi
      sequence: string
    - name: cn
      sequence: string
    - name: jp-yj
      sequence: string
    - name: ar
      sequence: string
    - name: es
      sequence: string
    - name: image
      dtype: image
  splits:
    - name: train
      num_bytes: 13852882321.001
      num_examples: 82783
    - name: val
      num_bytes: 811780220
      num_examples: 5000
    - name: restval
      num_bytes: 5123622277.68
      num_examples: 30504
    - name: test
      num_bytes: 823623386
      num_examples: 5000
  download_size: 20265033594
  dataset_size: 20611908204.681
configs:
  - config_name: default
    data_files:
      - split: train
        path: data/train-*
      - split: val
        path: data/val-*
      - split: restval
        path: data/restval-*
      - split: test
        path: data/test-*

Multilingual Common Objects in Context (COCO) Dataset

This dataset is a collection of multiple language open-source captions of COCO dataset.

The split in this dataset is set according to Andrej Karpathy's split from dataset_coco.json file. The collection was created specifically for simplicity of use in training and evaluation pipeline by non-commercial and research purposes. The COCO images dataset is licensed under a Creative Commons Attribution 4.0 License.

Multilanguage Feature's Code and Sources

If you use any part of the dataset, we recommend that you directly cite the original source for each language in this collection.

English en

English caption is retrieved from the original COCO dataset repository’s annotation file.

@misc{lin2015microsoftcococommonobjects,
      title={Microsoft COCO: Common Objects in Context}, 
      author={Tsung-Yi Lin and Michael Maire and Serge Belongie and Lubomir Bourdev and Ross Girshick and James Hays and Pietro Perona and Deva Ramanan and C. Lawrence Zitnick and Piotr Dollár},
      year={2015},
      eprint={1405.0312},
      archivePrefix={arXiv},
      primaryClass={cs.CV},
      url={https://arxiv.org/abs/1405.0312}, 
}

Thai th

Thai captions were a part of Romrawin Chumpu’s work at NECTEC. This work is partially supported by the Program Management Unit for Human Resources & Institutional Development, Research and Innovation (PMU-B) [Grant number B04G640107].
The captions were translated from English to Thai using google translate API.

Russian ru

Source: AlexWortega/ru_COCO: Translated coco dataset with "facebook/wmt19-en-ru" model
The captions were translated by using facebook/wmt19-en-ru model.

Japanese STAIR jp-stair

Source: STAIR Captions
The captions were translated from English to Japanese using machine translation.

@InProceedings{Yoshikawa2017,
  title     = {STAIR Captions: Constructing a Large-Scale Japanese Image Caption Dataset},
  booktitle = {Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers)},
  month     = {July},
  year      = {2017},
  address   = {Vancouver, Canada},
  publisher = {Association for Computational Linguistics},
  pages     = {417--421},
  url       = {http://aclweb.org/anthology/P17-2066}
}

Japanese YJ jp-yj

Source: yahoojapan/YJCaptions by Yahoo Japan.
Total captions of this Japanese version is around 26k captions.

Italian it

Source: crux82/mscoco-it: A large scale dataset for Image Captioning in Italian
The captions were obtained through semi-automatic translation from English to Italian.

German de

Source: Jotschi/coco-karpathy-opus-de · Datasets at Hugging Face
The captions were translated by using Helsinki-NLP/opus-mt-en-de · Hugging Face model.

Vietnamese vi

Source: dinhanhx/coco-2017-vi · Datasets at Hugging Face
The captions were translated by VinAI from English to Vietnamese.

@software{dinhanhx_VisualRoBERTa_2022,
    title        = {{VisualRoBERTa}},
    author       = {dinhanhx},
    year         = 2022,
    month        = 9,
    url          = {https://github.com/dinhanhx/VisualRoBERTa}
}

Chinese cn

Source: li-xirong/coco-cn: Enriching MS-COCO with Chinese sentences and tags for cross-lingual multimedia tasks
We selected only human generated dataset.

Arabic ar

Source: canesee-project/Arabic-COCO: MS COCO captions in Arabic
The captions were fully translated with Google's Advanced Cloud Translation API.

Spanish es

Source: carlosGarciaHe/MS-COCO-ES: MS-COCO-ES is a dataset created from the original MS-COCO dataset. This project aims to provide a small subset of the original image captions translated into Spanish by humans annotators. This subset is composed by 20,000 captions of 4,000 images.
The captions were translated by human.