iMaterialist Fashion Attribute (iFashion-Attribute)

iFashion-Attribute是由码隆科技等机构创建的大规模时尚图像数据集，包含超过100万张图像，涵盖228个细粒度时尚属性。该数据集由时尚专家精心标注，每张图像平均包含5个属性标签，旨在推动多标签和细粒度图像识别的研究。数据集的创建过程涉及从Wish平台随机抽样图像，并通过多轮后处理步骤提高标签质量。iFashion-Attribute特别适用于时尚领域的多标签分类和细粒度识别任务，为深度学习模型提供了丰富的训练资源，尤其是在解决时尚图像识别中的复杂问题方面具有重要价值。

iFashion-Attribute is a large-scale fashion image dataset created by Malong Technologies and other institutions, comprising over 1 million images spanning 228 fine-grained fashion attributes. Carefully annotated by fashion experts, each image contains an average of 5 attribute tags, and the dataset is designed to advance research in multi-label and fine-grained image recognition. The dataset construction process involves randomly sampling images from the Wish platform and implementing multi-round post-processing procedures to improve label quality. Specifically tailored for multi-label classification and fine-grained recognition tasks in the fashion domain, iFashion-Attribute provides abundant training resources for deep learning models, and holds significant value particularly in addressing complex challenges in fashion image recognition.