时尚视觉相似性基准数据集

时尚视觉相似性基准数据集是由开放大学等机构创建的大型数据集，包含超过110,000对由专家标注的图像对。该数据集主要用于评估视觉相似性发现方法，特别是在电子商务领域的应用。数据集中的图像对展示了不同对象间的高度视觉相似性，挑战了传统基于识别的评估方法。创建过程中，研究团队面临了标注效率和准确性的挑战，并提出了一种新的高效标注流程。该数据集的应用领域主要集中在改进视觉相似性模型，以提升搜索和推荐系统的性能。

The Fashion Visual Similarity Benchmark Dataset is a large-scale dataset developed by institutions including open universities, containing over 110,000 expert-annotated image pairs. It is primarily used to evaluate visual similarity discovery methods, especially for applications in the e-commerce domain. The image pairs in this dataset exhibit high visual similarity between distinct objects, which poses challenges to traditional recognition-based evaluation methods. During the dataset's creation, the research team faced challenges regarding annotation efficiency and accuracy, and proposed a novel efficient annotation workflow. The main application scenarios of this dataset focus on improving visual similarity models to enhance the performance of search and recommendation systems.