MIRFLICKR-25K

该数据集包含25,015张图片和223,635个标签，每张图片都与多个文本标签相关联，并标注有24维的语义标签。为了减少噪声，移除了出现次数少于20次的标签所对应的5,000张图片。数据集的规模为25,015张图片和223,635个标签，任务类型为跨模态检索。

This dataset contains 25,015 images and 223,635 labels. Each image is associated with multiple text labels and annotated with 24-dimensional semantic labels. To reduce noise, 5,000 images corresponding to labels that appeared fewer than 20 times were removed. The final dataset has a scale of 25,015 images and 223,635 labels, with the task type being cross-modal retrieval.

搜集汇总

背景与挑战

背景概述

MIRFLICKR-25K数据集包含25,015张图片和223,635个标签，每张图片关联多个文本标签并标注有24维语义标签，适用于跨模态检索任务。为减少噪声，已移除出现次数少于20次的标签对应的5,000张图片。