VQG
收藏arXiv2025-09-30 收录
下载链接：
https://www.microsoft.com/en-us/download/details.aspx?id=53670
下载链接
链接失效反馈官方服务：
资源简介：
该数据集包含了来自MSCOCO、Flickr和Bing的图片，每张图片都附有关于其自然且引人入胜的问题，大约有5000张图片，每张图片配备五个问题。需要注意的是，Bing图片集中有许多链接失效，因此在实验中并未考虑这部分数据。该数据集的规模大约为5000张图片，其任务是视觉问题生成。
This dataset includes images sourced from MSCOCO, Flickr, and Bing, with each image accompanied by natural and engaging questions related to its content. There are approximately 5,000 images in total, and each image is paired with five questions. It should be noted that a large number of links in the Bing image subset are invalid, so this portion of data was not utilized in experimental studies. The final valid scale of this dataset is approximately 5,000 images, and its core task is Visual Question Generation (VQG).
