Spatial and Temporal Understanding of Prepositions Dataset (STUPD)
收藏arXiv2023-09-13 更新2024-08-06 收录
下载链接：
http://arxiv.org/abs/2309.06680v1
下载链接
链接失效反馈官方服务：
资源简介：
STUPD是一个大规模的合成数据集，旨在帮助模型更好地理解静态和动态的空间关系，以及时间关系。数据集包含150,000个图像和视频，涵盖30种不同的空间关系，以及50,000个视频，涵盖10种时间关系。这些关系是从英语语言的介词中衍生出来的，通过Unity3D平台合成生成，包括详细的3D物体交互信息。STUPD数据集的目标是帮助模型在现实世界设置中更好地执行视觉关系检测。
STUPD is a large-scale synthetic dataset developed to enhance models' comprehension of static and dynamic spatial relationships as well as temporal relationships. The dataset contains 150,000 image and video samples covering 30 distinct spatial relationships, plus 50,000 video samples covering 10 different temporal relationships. All these relationships are derived from English prepositions, and are synthetically generated via the Unity3D platform, with detailed 3D object interaction metadata included. The core goal of the STUPD dataset is to facilitate better visual relationship detection performance by models in real-world scenarios.
提供机构：
新加坡科技研究局（A*STAR）
创建时间：
2023-09-13
