genrecon

Research code for GenRecon, a method that uses generative priors to improve 3D scene reconstruction from multiple views. It accompanies an academic paper.

Share on XLicense: MIT

Overview

GenRecon is the research code for a paper on high-fidelity 3D scene reconstruction from multi-view RGB images. It ties reconstruction closely to a strong generative 3D prior by treating scene reconstruction as conditional 3D generation over spatially localized, overlapping chunks that together tile the scene. This lets generation scale to large scenes. The authors are from the Technical University of Munich and Huawei Technologies Switzerland.

Key features

  • Multi-view RGB images to 3D scene reconstruction
  • Uses a generative 3D prior
  • Generates overlapping chunks that tile the scene

Best for

Researchers following work on generative priors for 3D reconstruction who want to read or run the paper's code. The README excerpt gives no install details.

Upstream
kasothaphie/GenRecon
Fork on GitHub
Guo-astro/genrecon
Upstream stars
1.0k
Category
Finance, data and research
Language
Python
License
MIT
Forked
2026-06-09
Sync status
In syncLast synced 2026-09-29