数据堂—700,000组通用场景图像描述数据（简单描述）

700,000组图像及描述，图片类型涵盖风景、动物、花卉树木、人物、汽车、运动、工业以及建筑等多种类别及一个美学子集，每张图不少于两种描述，每种描述一句话；少部分图像只有一种描述，描述语言为英语，中文两种语言

This dataset contains 700,000 image-caption pairs. The images span diverse categories including landscapes, animals, flowers and trees, human figures, automobiles, sports scenes, industrial scenarios, architecture, as well as a subset focused on American students. Each image is paired with at least two captions, with each caption consisting of one single sentence; a small number of images only have one caption. All captions are available in both English and Chinese.

搜集汇总

背景与挑战

背景概述

该数据集包含70万组通用场景图像及其描述，覆盖风景、动物、人物等多个类别，每张图像至少有两个中英文句子描述。它主要用于测试通用场景图像描述任务，数据格式为.jpg图像和.txt文本，描述准确率超过95%。