ViVQA-X / README.md
duongtruongbinh's picture
Update README.md
1eb23c6 verified
metadata
configs:
  - config_name: default
    data_files:
      - split: train
        path: ViVQA-X_train.json
      - split: test
        path: ViVQA-X_test.json
      - split: val
        path: ViVQA-X_val.json
task_categories:
  - visual-question-answering
language:
  - vi
pretty_name: ViVQA-X
size_categories:
  - 10K<n<100K
tags:
  - visual-question-answering
  - Vietnamese
  - machine-translation
  - evaluation
  - LLM
annotators:
  - Automated Translation and Evaluation Pipeline

Dataset Card for ViVQA-X

Dataset Description

ViVQA-X is the Vietnamese version of the VQA-X dataset, designed for tasks in VQA-NLE in the Vietnamese language. This dataset was created to support research in Vietnamese VQA, offering translated examples with explanations, in alignment with the original English dataset. The ViVQA-X dataset is generated using a multi-stage pipeline for translation and evaluation, including both automated translation and post-processing, with quality evaluations involving multiple large language models (LLMs).

This dataset includes annotation files that describe the questions, answers, and explanations. The corresponding images used in the dataset are sourced from the original COCO dataset. You can download the COCO images from the official COCO website.

Dataset statistics

  • QA Pairs: 32,886 pairs of questions and answers across Train, Validation, and Test splits.
  • Explanations: 41,817 high-quality explanations.
  • Average Words: Explanations average 10 words per explanation.
  • Vocabulary Size: 4,232 unique words in explanations.

JSON Structure and Field Explanations

Each JSON object in the dataset follows this structure similar to the original VQA-X dataset:

{
  "question": "Đây là phòng nào?",
  "image_id": "524822",
  "image_name": "COCO_val2014_000000524822.jpg",
  "explanation": [
    "nó là một phòng có ghế sofa và tivi",
    "căn phòng này có một lò sưởi, một chiếc ghế dài, nhiều ghế bành và một chiếc tivi gắn trên tường",
    "có một chiếc ghế sofa kiểu chữ L và một chiếc tivi và lò sưởi"
  ],
  "answer": "phòng khách",
  "question_id": "524822007"
}
  • question: The question asked about the image.
  • image_id: The ID of the image from the original COCO dataset.
  • image_name: The file name of the COCO image used to reference the specific image.
  • explanation: A list of explanations.
  • answer: The answer to the question (the translation of the most common answer in the original dataset).
  • question_id: The ID of the question.

Visual Examples

Below are examples that include the image, question, answer, and explanations.

Example 1

COCO_val2014_000000524822

  • Question: Đây là phòng nào?
  • Answer: phòng khách
  • Explanations:
    • nó là một phòng có ghế sofa và tivi
    • căn phòng này có một lò sưởi, một chiếc ghế dài, nhiều ghế bành và một chiếc tivi gắn trên tường
    • có một chiếc ghế sofa kiểu chữ L và một chiếc tivi và lò sưởi

Example 2

COCO_val2014_000000264279

  • Question: Anh ấy đang chơi môn thể thao gì?
  • Answer: lướt sóng
  • Explanations:
    • người đó đang đứng trên một tấm ván ở biển
    • anh ấy đang đứng trên một tấm ván trên mặt nước có một con sóng nhỏ
    • anh ấy đang ở trên đỉnh sóng

If you mention or use any information from our dataset, please cite our paper:

@misc{vivqax2025,
  author    = {Duong, Truong-Binh and Tran, Hoang-Minh and Le-Nguyen, Binh-Nam and Duong, Dinh-Thang},
  title     = {An Automated Pipeline for Constructing a Vietnamese VQA-NLE Dataset},
  howpublished = {Accepted for publication in the Proceedings of The International Conference on Intelligent Systems \& Networks (ICISN 2025), Springer Lecture Notes in Networks and Systems (LNNS)},
  year      = {2025}
}