ViOCRVQA: Vietnamese Optical Character Recognition - Visual Question Answering

ViOCRVQA Dataset

Welcome to ViOCRVQA (Vietnamese Optical Character Recognition - Visual Question Answering ) dataset! This dataset is the largest scale dataset in Vietnamese specializing in understanding text appearing in images.

Overview

ViOCRVQA contains over consisting of 28,000+ images and 120,000+ question-answer pairs. In this dataset, all the images contain text and questions about the information relevant to the text in the images.

Purpose

The purpose of ViOCRVQA is to provide a benchmark for evaluating the reading comprehension ability of Visual Question Answering (VQA) models in the Vietnamese language. As a developing country, Vietnam is still in need of resources and benchmarks to advance research in AI and machine learning.

Key Features

28,282 images
123,781 questions with answers
Focus on understanding text within images
Meticulously crafted to ensure diverse and challenging questions

Importance of ViOCRVQA

Understanding text in images is crucial for many real-world applications, such as assisting the visually impaired, enhancing image search engines, and improving AI understanding of multimedia content. convenient. ViOCRVQA fills an important gap by providing the largest-scale dataset relevant to Vietnamese.

Usage

Researchers and developers can use ViOCRVQA to train and evaluate their VQA models, analyze the performance of different approaches, and contribute to advancing research in this field. The dataset is freely available for research purposes.

Contributions

Create the largest-scale dataset for text-based VQA tasks in Vietnamese, focusing on text appearing in images.
Analyze the challenge of the ViOCRVQA dataset by evaluating the performance of the OCR system.
Through extensive testing, we found that the VQA models used for English are not really effective on Vietnamese. We recommend our proposed VisionReader model
Our experiments demonstrate the effectiveness of building relationships between objects and text information in images.

Availability

You can access it here: ViOCRVQA Dataset.

Citation

If you use ViOCRVQA dataset in your research, please cite our paper:

Citation

If you use the ViOCRVQA dataset in your research, please cite our paper:

@article{pham2025viocrvqa,
  title     = {ViOCRVQA: novel benchmark dataset and VisionReader for visual question answering by understanding Vietnamese text in images},
  author    = {Pham, Huy Quang and Nguyen, Thang Kien-Bao and Van Nguyen, Quan and Tran, Dan Quang and Nguyen, Nghia Hieu and Van Nguyen, Kiet and Nguyen, Ngan Luu-Thuy},
  journal   = {Multimedia Systems},
  volume    = {31},
  number    = {2},
  pages     = {106},
  year      = {2025},
  publisher = {Springer}
}

Authors

Huy Quang Pham
- Email: [email protected]
Thang Kien-Bao Nguyen
- Email: [email protected]
Quan Van Nguyen
- Email: [email protected]
Dan Quang Tran
- Email: [email protected]
BS Nghia Hieu Nguyen
- Email: [email protected]
MSc Kiet Van Nguyen
- Email: [email protected]
Assoc. Prof Ngan Luu-Thuy Nguyen
- Email: [email protected]

Affiliations

Faculty of Information Science and Engineering, University of Information Technology
Vietnam National University, Ho Chi Minh City, Vietnam

Contact

For any inquiries or feedback regarding the ViOCRVQA dataset, please contact [email protected].

Thank you for your interest in ViOCRVQA! We hope this dataset contributes to the advancement of research in text-based Visual Question Answering around the world, especially in Vietnam.