JTA*

JTA*数据集是由韩国蔚山科学技术院人工智能研究生院创建，用于训练人物搜索网络。该数据集包含约450,000张从《侠盗猎车手V》游戏中提取的图像，每张图像均自动标注了人物实例的边界框、身份和关键点。JTA*数据集通过选择训练类别中的256个序列，进一步构建为JTA*，用于人物搜索任务。数据集的特点在于其自动标注的准确性，以及多样化的背景、视角和天气条件，有助于提高网络在真实数据集上的泛化能力。该数据集主要应用于人物搜索领域，旨在解决大规模数据集收集和标注的时间消耗、劳动密集及隐私问题。

The JTA* dataset was developed by the Graduate School of Artificial Intelligence at the Ulsan National Institute of Science and Technology (UNIST), Republic of Korea, for training person search networks. This dataset contains approximately 450,000 images extracted from the video game *Grand Theft Auto V*, with each image automatically annotated with bounding boxes, identities, and keypoints of person instances. To build a task-specific variant for person search, the JTA* dataset is further refined by selecting 256 sequences from its training categories. The dataset is distinguished by its highly accurate automatic annotations, as well as diverse backgrounds, viewing perspectives, and weather conditions, which help improve the generalization performance of networks on real-world datasets. This dataset is primarily applied in the field of person search, aiming to address the issues of time consumption, labor intensity, and privacy concerns associated with collecting and annotating large-scale datasets.