ITCPR dataset

该数据集名为ITCPR，包含了2,225个标注的三元组，用于组合人物检索任务，涵盖了2,202种独特的查询图像和文本注释组合。该数据集是从公开可获得的换衣数据集中构建的，如Celeb-reid、PRCC和LAST，确保图像展示了同一人在不同服装或场景下的样子。数据规模包括2,225个三元组、1,151张图像和512个身份。其任务是零样本组合人物检索（Zs-Cpr）。The dataset named ITCPR contains 2,225 annotated triplets for compositional person retrieval tasks, covering 2,202 unique query image-text annotation pairs. It is constructed from publicly available clothing-changing datasets including Celeb-reid, PRCC and LAST, ensuring that the images depict the same individual wearing different clothing or in different scenarios. The dataset consists of 2,225 triplets, 1,151 images and 512 unique identities, with the target task being Zero-shot Compositional Person Retrieval (Zs-Cpr).