Proceedings of the Annual Meeting of the Cognitive Science Society

Human physical scene understanding requires more than simply localizing and recognizing objects Ñ we can quickly adapt our predictions about how a scene will unfold by incorporating objects' latent physics properties, such as the masses of the objects in the scene. What are the underlying computational mechanisms that allow humans to infer these physical properties and adapt their physical predictions so efficiently from visual inputs? One hypothesis is that general intuitive physics knowledge can be learned from enough raw data, instantiated as computational models that predict future video frames in large datasets of complex scenes. To test this hypothesis, we evaluate existing state-of-the-art video models. We measured both model and human performance on Physion++, a novel dataset and benchmark that rigorously evaluates visual physical prediction in humans and machines, under circumstances where accurate physical prediction relies on accurate estimates of the latent physical properties of objects in the scene. Specifically, we tested scenarios where accurate prediction relied on accurate estimates of objects' mechanical properties, including masses, friction, elasticity and deformability, and the values of these mechanical properties could only be inferred by observing how these objects moved and interacted with other objects and/or fluids. We found that models that encode objectness and physical states tend to perform better, yet there is still a huge gap compared to human performance. We also found most models' predictions correlate poorly with that made by humans. These 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.