NUS-WIDE (Web)

该数据集名为NUS-WIDE（网络版），旨在用于多标签表示学习，包含了众多类别及其相关的图片。此外，该数据集还用于评估在噪声加权自监督学习设置下，CAPro的鲁棒性。其所涉及的任务是多项标签表示学习。

This dataset, named NUS-WIDE (Web Edition), is designed for multi-label representation learning and contains numerous categories and their associated images. Additionally, this dataset is also used to evaluate the robustness of CAPro under the noise-weighted self-supervised learning setting. The core task involved in this dataset is multi-label representation learning.

搜集汇总

背景与挑战

背景概述

NUS-WIDE (Web)是一个大规模网络图像数据集，包含81个概念标签，这些标签是从用户标签中提取的弱标签，具有现实噪声，适用于多标签场景下的网络监督学习研究。数据集提供了官方的训练和测试分割，并支持TF-Record格式以优化I/O性能。