Phys4D: Fine-Grained Physics-Consistent 4D Modeling from Video Diffusion
技术动态
来源：arXiv cs.RO发布时间待核实
arXiv:2603.03485v4 Announce Type: replace-cross Abstract: Recent video diffusion models have achieved impressive capabilities as large-scale generative world models. However, these models often struggle with fine-grained physical consistency, exhibiting physically implausible dynamics over time. In this work, we present \textbf{Phys4D}, a pipeline for learning physics-consistent 4D world representations from video diffusion models.