MALS

MALS是一个大规模的多属性与语言搜索数据集，专为文本基础人物检索设计。该数据集包含约1,510,330对图像-文本，是现有数据集CUHK-PEDES的37.5倍。所有图像均标注了27种属性，如性别、年龄、发型等。MALS数据集通过利用扩散模型自动生成，有效解决了隐私保护和标注成本高的问题。数据集的创建旨在提供一个丰富、多样化的资源，以支持文本基础人物检索的研究和应用，特别是在提高模型对人物特征的理解和匹配精度方面。

MALS is a large-scale multi-attribute and language search dataset specifically designed for text-based person retrieval. It contains approximately 1,510,330 image-text pairs, which is 37.5 times the size of the existing CUHK-PEDES dataset. All images are annotated with 27 attributes including gender, age, hairstyle and others. The MALS dataset is automatically generated via diffusion models, effectively addressing the challenges of privacy protection and high annotation costs. This dataset is developed to provide a rich and diverse resource to support the research and application of text-based person retrieval, particularly in enhancing the model's understanding of person characteristics and improving matching accuracy.