CxC
收藏arXiv2025-09-30 收录
下载链接：
https://github.com/google-research-datasets/crisscrossed-captions
下载链接
链接失效反馈官方服务：
资源简介：
该数据集在MS-COCO的基础上进行了扩展，加入了由人类标注的语义相似度评分，范围从0到5，涵盖了图像与图像、描述与描述以及图像与描述之间的配对。在实验中，我们从图像与图像以及图像与描述的数据集中各选取了1000对平衡子集。该数据集的任务是对语义相似度进行评估。
This dataset is extended based on the MS-COCO dataset, and incorporates human-annotated semantic similarity scores ranging from 0 to 5, covering pairs of image-image, caption-caption, and image-caption. In experiments, we selected a balanced subset of 1000 pairs respectively from the image-image and image-caption datasets within this corpus. The task of this dataset is semantic similarity evaluation.
