Reproduce “CityLift: Compositional 3D Urban Generation from a Single Image” 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

Haoran Gong, Jie Feng, Chenxu Wang

CityLift generates complete, textured, editable building meshes from an oblique urban image and instance masks. It builds a shared block scaffold, transfers each building’s spatial support and scene-conditioned features into local generation, then returns the meshes to their scene positions. On held-out CityLift-5K, scene-level Chamfer distance is 1.079 grid cells, versus 2.595 for SAM 3D; inputs require masks.

Summary by alphaXiv.

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