OID-IOTA-10K

该数据集是从开放图像数据集中新收集的一组包含1万个图像的“信息丰富”的真实标签。每张图像都由三名评分员进行标注，重点关注图像中的主要物体或场景。此外，数据集还包括与预定义知识图谱匹配的自由文本标签，并设有处理模糊性和同义词的过程。规模涵盖1万张图像，任务涉及图像标注和标签排名。

This dataset is a newly curated set of 10,000 "information-rich" ground-truth labels for images, sourced from the Open Images Dataset. Each image is annotated by three annotators, with the annotation focusing on the primary objects or scenes depicted in the image. Furthermore, the dataset includes free-text labels that match a pre-defined knowledge graph, and integrates a structured procedure for handling ambiguity and synonymy. Comprising a total of 10,000 images, the associated tasks cover image annotation and label ranking.