VQA Dataset
收藏知名数据库2026-06-26 收录
下载链接：
https://visualqa.org
下载链接
链接失效反馈官方服务：
资源简介：
VQA是一个视觉问答数据集，包含关于图像的开放式问题，需要结合视觉、语言和常识知识来回答。该数据集包含265,016张图像(COCO和抽象场景)，每张图像平均有5.4个问题，每个问题有10个标准答案和3个 plausible 但可能不正确的答案。数据集于2017年首次完整发布，并于2025年获得Mark Everingham Prize，以表彰其对计算机视觉社区的贡献。
VQA is a visual question answering dataset featuring open-ended questions about images, which demand the integration of visual, linguistic and common-sense knowledge to answer correctly. The dataset comprises 265,016 images sourced from COCO and abstract scenes, with an average of 5.4 questions per image. Each question is paired with 10 standard answer options and 3 plausible yet potentially incorrect alternatives. The dataset was first fully released in 2017, and was awarded the Mark Everingham Prize in 2025 in recognition of its contributions to the computer vision community.
