The ability to understand physical dynamics is critical for agents to act in the world. Here, we use Counterfactual World Modeling (CWM) to extract vision structures for dynamics understanding. CWM uses a temporally-factored masking policy for masked prediction of video data without annotations. This policy enables highly effective "counterfactual prompting" of the predictor, allowing a spectrum of visual structures to be extracted from a single pre-trained predictor without finetuning on annotated datasets. We demonstrate that these structures are useful for physical dynamics understanding, allowing CWM to achieve the state-of-the-art performance on the Physion benchmark.
@article{arxiv.2312.06721,
title = {Understanding Physical Dynamics with Counterfactual World Modeling},
author = {Rahul Venkatesh and Honglin Chen and Kevin Feigelis and Daniel M. Bear and Khaled Jedoui and Klemen Kotar and Felix Binder and Wanhee Lee and Sherry Liu and Kevin A. Smith and Judith E. Fan and Daniel L. K. Yamins},
journal= {arXiv preprint arXiv:2312.06721},
year = {2024}
}