Hugging Face's logo

Datasets:
minhquan6203
/
ViTextVQA

ArXiv:
Dataset card Files Community
1

You need to agree to share your contact information to access this dataset

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

Log in or Sign Up to review the conditions and access this dataset content.

YAML Metadata Warning: empty or missing yaml metadata in repo card (https://huggingface.co/docs/hub/datasets-cards)

ViTextVQA: A Large-Scale Visual Question Answering Dataset for Evaluating Vietnamese Text Comprehension in Images

examples

ViTextVQA Dataset

Welcome to ViTextVQA (Vietnamese Text-based Visual Question Answering) dataset! This dataset is the first high-quality large-scale dataset in Vietnamese specializing in understanding text appearing in images.

Overview

ViTextVQA contains over 16,000 images and over 50,000 questions with answers. The dataset is designed to evaluate the ability of AI models to comprehend text within images and answer questions based on that understanding.

Purpose

The purpose of ViTextVQA 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

  • 16,762 images
  • 50,342 questions with answers
  • Focus on understanding text within images
  • Meticulously crafted to ensure diverse and challenging questions

Importance of ViTextVQA

Understanding text in images is crucial for many real-world applications, such as assisting visually impaired individuals, enhancing image search engines, and improving AI's understanding of multimedia content. ViTextVQA fills a crucial gap by providing a large-scale dataset tailored to the Vietnamese language.

Usage

Researchers and developers can use ViTextVQA 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 first high-quality large-scale dataset for text-based VQA task in Vietnamese, focusing on scene text and text appearing in the image.

  • Analyze the challenge of the ViTextVQA dataset by evaluating the performance of the OCR system.

  • Through our extensive experiments, we found that VQA models using ViT5 as their backbone behave as the answer selector methods when OCR text is suffixed for the question.

  • Our experiments showed the effectiveness of arranging from top-left to bottom-right, resulting in remarkable enhancements in the performance.

Availability

The ViTextVQA dataset will be available for download after our article is accepted.

You can find it at the following link: ViTextVQA Dataset

Evaluation

Note that you should combine both dev and test files and submit them on Kaggle to get the most accurate evaluation results.

Citation

If you use ViTextVQA dataset in your research, please cite our paper (preprint):

ViTextVQA: A Large-Scale Visual Question Answering Dataset for Evaluating Vietnamese Text Comprehension in Images

Authors

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 ViTextVQA dataset, please contact 21521333@gm.uit.edu.vn or haryquan.minh@gmail.com.

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

April 15, 2024

Downloads last month
373
Edit dataset card