Reproduce “GeoV2V: Geometry-Grounded Video Diffusion Model for Driving Scene 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

Yuchen Xi, Yuhan Guo, Chenwei Hou, Zixu Liu, Jin Fang, Jason Liu, Ruigang Yang

GeoV2V generates views from alternate driving lanes by combining source-video appearance with geometric cues that guide the requested viewpoint. Para4D supplies synchronized simulated views from parallel lanes for training. For a 2-meter real-world lane shift, GeoV2V reports an FVD of 232.06 versus StreetCrafter’s 272.85, and the lowest FVD across tested offsets, though its FID is not best at 2 or 4 meters.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.