Reproduce “SyncVid: World-Equivariant Diffusion for Synchronized 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
Farhad G. Zanjani, Animesh Karnewar, Fatih Porikli, Amirhossein Habibian
SyncVid makes synchronized multi-camera video reshooting a property of diffusion’s starting noise. It attaches persistent random vectors to canonical anchors in a reconstructed dynamic 3D scene and renders them into target views, with approximate whitening to reduce local noise correlations. On Multi-ViewBench, it achieves FVD-V 146.99 versus 202.22 for Reshoot-Anything, but these proxy scores do not establish exact target-view correctness.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.