Reproduce “PhysOPSD: Code as Privilege for Physically Plausible Video Generation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Harold Haodong Chen, Rui Peng, Lingteng Qiu, Kejie Qiu, Chao Xu, Zilong Dong
PhysOPSD trains video generators to represent physical motion using executable simulations as privileged training context. An exponential-moving-average teacher sees detailed and simplified drafts, then supervises student-generated rollouts toward text-only generation; controlled scene pairs vary one physical parameter. On UniVideo, VideoPhy-2 All rises from 63.1 to 67.3 with 3K post-training samples, and VBench Total is 83.5 versus 84.0.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.