CUB-DG

此数据集包含200类别的北美鸟类的整体11,788图像。先前已收集了每个图像的十个句子 [33]，其中提供了图像内容的详细描述，例如，“这只鸟有一个长而尖的钞票，白色的腹部和黑色的皇冠。” 这个数据集一直是视觉解释任务的理想基准，因为句子是特定于类的和区分类的。This dataset contains a total of 11,788 images across 200 categories of North American birds. Ten sentences have been previously collected for each image [33], which provide detailed descriptions of the image content. For example, "This bird has a long, pointed beak, white belly and black crown." This dataset has served as an ideal benchmark for visual explanation tasks, as the sentences are category-specific and capable of distinguishing different bird categories.

搜集汇总

背景与挑战

背景概述

CUB-DG数据集包含200类北美鸟类的11,788张图像，每张图像配有十个详细描述的句子，适用于视觉解释任务。该数据集由首尔国立大学、高丽大学和LG AI Research于2022年发布。