ComPhy

该数据集是一个专注于理解那些从视觉外观中隐藏的以对象为中心和关系性物理属性的视觉问答数据集。它研究物体交互中固有的物理属性，以及这些属性如何影响未来的运动。该数据集的任务是视频问答。

This dataset is a visual question answering (VQA) dataset focused on understanding object-centric and relational physical properties that are hidden from visual appearances. It investigates the inherent physical properties underlying object interactions, as well as how these properties impact subsequent motion. The task associated with this dataset is video question answering (VideoQA).