Reproduce “GeoPano: Towards Geometrically Accurate 3D Reconstruction from a Single Panorama” 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

Jing OU, Zidong Cao, Liaoyuan Fan, Zhuoxiao Li, Shuai Zhang, Yinrui Ren, Dongli Wu, Haoang Li and 2 more

GeoPano reconstructs a fused point cloud from one panorama by decomposing it into overlapping perspective views and restricting cross-view feature exchange according to spherical geometry. It combines local point-cloud supervision with a panoramic loss to balance detail and global alignment. On Matterport3D, its Chamfer distance is 0.153 versus 0.337 for RPG360, but the methods use different training paradigms.

Summary by alphaXiv.

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