Clothing1M

该数据集名为Clothing-1M，包含了超过一百万张256x256分辨率的服装图片，分为14个类别。它专为带有噪声标签的图像识别而设计，并已成为图像识别领域中带有噪声标签的最广泛认可的基准测试。该数据集规模庞大，拥有超过一百万张图片，主要任务为图像分类。

The dataset named Clothing-1M, which contains over one million clothing images with a resolution of 256×256 and is categorized into 14 classes, is specifically designed for image recognition with noisy labels. It has become the most widely recognized benchmark for noisy-label learning in the field of image recognition, and as a large-scale dataset with over one million images, its core task is image classification.

搜集汇总

背景与挑战

背景概述

Clothing1M是一个大规模图像分类数据集，包含干净标签和噪声标签，旨在研究从噪声标签中学习有效模型的方法。数据集结构包括图像文件夹和多个标签文件，支持多种实验设置和模型训练。