OV-VG

OV-VG数据集由北京航空航天大学电子信息工程学院的研究团队创建，旨在支持开放词汇视觉定位（OV-VG）研究。该数据集包含7272张从MS COCO选取的图像，涵盖10000个实例，分为基础类别和新颖类别，以促进对未预定义词汇中概念的理解。数据集的构建过程包括精细的类别选择和专业的标注策略，确保了数据的高质量和适用性。OV-VG数据集特别关注视觉与语言信息的精确对齐，适用于多种实际应用，如机器人导航和视觉对话，旨在解决复杂场景下的目标定位问题。

The OV-VG dataset was developed by the research team from the School of Electronic and Information Engineering, Beihang University, to support research on open-vocabulary visual grounding (OV-VG). It comprises 7272 images selected from MS COCO, covering 10,000 instances, and is categorized into base classes and novel classes, aiming to facilitate the comprehension of concepts in non-predefined vocabularies. The construction of the OV-VG dataset involves meticulous category selection and professional annotation strategies, which ensures high data quality and applicability. The OV-VG dataset specifically focuses on precise alignment between visual and linguistic information, and is applicable to various practical applications such as robot navigation and visual dialogue, with the goal of solving target localization problems in complex scenarios.