NegVQA
收藏arXiv2025-09-30 收录
下载链接：
https://yuhui-zh15.github.io/NegVQA/
下载链接
链接失效反馈官方服务：
资源简介：
该数据集名为NegVQA，是一个视觉问答（VQA）的基准，包含了7,379个两选题，覆盖了各种否定情境以及图像和问题的分布。该数据集旨在评估视觉语言模型在不同视觉场景下理解否定的能力，确保在否定形式的多样性以及与视觉内容的强烈相关性方面具有广泛的覆盖面。该数据集的规模为7,379个问题，所涉及的任务是视觉问答。
NegVQA is a Visual Question Answering (VQA) benchmark dataset. It includes 7,379 binary-choice questions, covering diverse negation scenarios and distributions of images and their associated questions. This benchmark is designed to assess the ability of vision-language models to comprehend negation across various visual scenarios, ensuring extensive coverage of diverse negation forms and strong correlations with visual content. With 7,379 questions in total, its core task is Visual Question Answering.
提供机构：
Constructed by the authors using GPT-4o
