---
language:
- vi
- en
license: cc-by-4.0
size_categories:
- 100k<n<1M
task_categories:
- image-to-text
- text-to-image
- translation
tags:
- vision
- image-captioning
- ms-coco
- coco
- vietnamese
- vietnam
- vi
- vn
homepage: https://aienthusiasm.vn
dataset_info:
  features:
  - name: image_id
    dtype: string
  - name: caption_id
    dtype: string
  - name: image
    dtype: image
  - name: caption_en
    dtype: string
  - name: caption_vi
    dtype: string
  splits:
  - name: train
    num_bytes: 95104738954.827
    num_examples: 591753
  - name: validation
    num_bytes: 3666417387.754
    num_examples: 25014
  download_size: 67774085459
  dataset_size: 98771156342.581
configs:
- config_name: default
  data_files:
  - split: train
    path: data/train-*
  - split: validation
    path: data/validation-*
---
## Team and Homepage
This dataset is maintained by **AI Enthusiasm**. You can find more of our work, community projects, and official updates at our homepage:
- **Official Website**: [https://aienthusiasm.vn](https://aienthusiasm.vn)
- **Hugging Face Organization**: [https://huggingface.co/ai-enthusiasm-community](https://huggingface.co/ai-enthusiasm-community)

## Contact
If you encounter any issues with the dataset or have any inquiries, please feel free to reach out to us via email at: [aienthusiasm.team@gmail.com](mailto:aienthusiasm.team@gmail.com)

## Dataset Summary
COCO-2017-Vietnamese is a localized version of the Microsoft Common Objects in Context (COCO) 2017 dataset, a large-scale object detection, segmentation, and captioning dataset. This version is specifically curated for Vietnamese cross-modal research, featuring the original English captions paired with high-quality Vietnamese translations. It serves as a comprehensive benchmark for tasks such as Image Captioning and Multimodal Learning within a bilingual framework.

## Dataset Structure
The dataset is provided in a flattened tabular format, optimized for the Hugging Face Dataset Viewer and high-speed Parquet processing.

### Data Instances
Each instance represents a single image-caption pair. To maintain compatibility with standard training pipelines, image data is repeated for each associated caption.

### Data Fields
- `image_id`: The ID of the image from the original COCO dataset.
- `caption_id`: The unique ID for each specific annotation (caption).
- `image`: An Image object containing the visual data.
- `caption_en`: The original descriptive text in English.
- `caption_vi`: The translated descriptive text in Vietnamese.

## Usage
The dataset can be accessed directly using the Hugging Face `datasets` library:

```python
from datasets import load_dataset

dataset = load_dataset("ai-enthusiasm-community/coco-2017-vietnamese")

# Accessing the first sample from the training set
print(dataset['train'][0])
```

## Citation
```
@inproceedings{lin2014microsoft,
  title={Microsoft coco: Common objects in context},
  author={Lin, Tsung-Yi and Maire, Michael and Belongie, Serge and Hays, James and Perona, Pietro and Ramanan, Deva and Doll{'a}r, Piotr and Zitnick, C Lawrence},
  booktitle={Computer Vision--ECCV 2014: 13th European Conference, Zurich, Switzerland, September 6-12, 2014, Proceedings, Part V 13},
  pages={740--755},
  year={2014},
  organization={Springer}
}
```