CC3M-QA-DC
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/johncaged/OPT_Questioner
下载链接
链接失效反馈官方服务：
资源简介：
该数据集是从概念性标题（CC3M）数据集中生成的新数据集，它包含了用于视觉问题解答和密集注释的图像-问题-答案以及图像-位置-标题的三元组。当以多任务方式进行预训练时，该数据集能够提升在各类下游任务上的性能。其所涉及的任务包括视觉问题解答和密集注释。
This new dataset is generated from the Conceptual Captions (CC3M) dataset, and it contains image-question-answer and image-location-caption triples for visual question answering (VQA) and dense captioning tasks. When pre-trained in a multi-task manner, this dataset can improve performance across various downstream tasks. The tasks involved include visual question answering and dense captioning.
提供机构：
Authors of the paper
