Building Rome from a single image: new method reconstructs 3D scenes with geometry beyond the visible view
jonstephens85 · x · 2026-10-09
Researchers from Applied Intuition, Purdue, UIUC and UC Berkeley released "Building Rome from a Single Image," a method that reconstructs full 3D scenes from one photo, including geometry outside the observed view, across indoor, outdoor and large-scale settings—without generating fake views and running SfM.
Key ideas:
- Adaptive scene chunks that grow with distance: small chunks preserve near-field detail, large ones cover distant buildings
- Explicit 2D-3D conditioning: image features are lifted onto observed surfaces, and each voxel knows whether it is free, visible, or hidden
- Autoregressive generation: neighboring chunks reuse overlapping latents to keep new geometry consistent
Compared against World Tracing, GenRecon, VolFill and Lyra 2.0. Paper on arXiv, code coming soon, plus an interactive web viewer with orbit/zoom and point-to-point metric measurement (meshes simplified to 1.5M faces).
Related event: Building Rome: Reconstructing Full 3D Scenes from a Single Image(2 posts)→
More from Multimodal
- Epic ships official Unreal MCP, letting Claude Code and Cursor drive the Unreal Editor — Scobleizer · 2026-10-09
- ByteDance's Dreamina launches AI film label with up to 15M credits and $10K promotion per project — lmoroney · 2026-10-09
- Score Studio runs a "leaked" GTA 6 dirt race, catching every overtake and dust plume — markjeffrey · 2026-10-09
- LightOnOCR-3 Seamlessly Extracts Hard-to-Read German Text in Demo — IgorCarron · 2026-10-09
- LightOnOCR-2-1B: a 1B-parameter open OCR model hits SOTA at under $0.01 per 1,000 pages — IgorCarron · 2026-10-09
- Open-source local image library asks: how do you find an old generation by its settings? — shivam_dewan · 2026-10-09