Physics-IQ

该数据集是一个全面性的基准数据集，旨在评估生成视频模型是否学会了物理原理，如流体动力学、光学、固体力学、磁学和热力学等。该数据集包含了多种场景的录制，每个场景都录制了两次，以估算现实世界中物理现象的自然变异性。该数据集的任务是理解和量化生成视频模型所学习的物理原理。

This is a comprehensive benchmark dataset designed to evaluate whether video generation models have mastered physical principles including fluid dynamics, optics, solid mechanics, magnetism, and thermodynamics. The dataset contains recordings of multiple scenarios, with each scenario being recorded twice to estimate the natural variability of physical phenomena in the real world. The task of this dataset is to understand and quantify the physical principles learned by video generation models.