WuDaoMM

一个包含超过6.5亿对图像-文本配对的大型多模态语料库，其中约6亿对数据来自图像和标题弱相关的多个网页，另外5000万对强相关的图像-文本配对来自一些高质量的图形网站。

A large-scale multimodal corpus containing over 650 million image-text pairs, of which approximately 600 million pairs are sourced from multiple webpages where images and their associated titles are weakly correlated, while the remaining 50 million strongly correlated image-text pairs are obtained from several high-quality graphic websites.

搜集汇总

背景与挑战

背景概述

WuDaoMM是一个大规模多模态数据集，包含超过6.5亿对图像-文本配对。其中，约6亿对数据来自网页，图像和标题弱相关；另外5000万对来自高质量图形网站，图像和文本强相关。该数据集以大规模和多样化数据来源为特点，适用于多模态学习和相关研究。