Reproduce “Voyaging into Perpetual Dynamic Scenes from a Single View” 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
Fengrui Tian, Tianjiao Ding, Jinqi Luo, Hancheng Min, René Vidal
Researchers from the University of Pennsylvania developed DynamicVoyager, a method that generates perpetual, dynamic 3D scenes from a single initial view by iteratively outpainting new dynamic content while maintaining 3D consistency. This approach enables exploration along unbounded camera trajectories through living virtual worlds, outperforming prior static-only and bounded dynamic scene generation techniques.
Key insights
- Reformulating perpetual dynamic scene generation as an iterative scene outpainting problem with new dynamic content is essential for overcoming boundedness.
- Enriching 2D video outpainting with 3D ray information provides crucial geometric context, allowing 2D generative models to maintain 3D consistency for large camera movements.
- A structured 4D representation that distinguishes dynamic foregrounds from static backgrounds and handles background completion effectively contributes to overall scene consistency and realism.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.