Physion++ is a novel dataset and benchmark designed to evaluate visual physical prediction in artificial systems under circumstances where those predictions rely on accurate estimates of the latent physical properties of objects in the scene. The benchmark tests scenarios where accurate prediction relies on estimates of properties such as mass, friction, elasticity, and deformability, and where the values of those properties can only be inferred by observing how objects move and interact with other objects or fluids. The benchmark is designed to evaluate the performance of a variety of state-of-the-art prediction models and compare that performance to a set of human predictions. The results show that current deep learning models that succeed in some settings nevertheless fail to achieve human-level physical prediction in other cases, especially those where latent property inference is required.

Publication date: June 27, 2023
Project Page: https://dingmyu.github.io/physion_v2/
Paper: https://arxiv.org/pdf/2306.15668.pdf