Reproduce “Deep Bundle Adjustment” 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

Pak Hong Chui, Hector Andrade-Loarca, Daniel Cremers

DeepBA learns an iterative solver for multi-view 3D reconstruction, converting pairwise image matches into camera poses, intrinsics, and depth from a trivial starting state without building tracks. On RealEstate10K, it scores 84.0 AUC@30°, versus GASfM’s 82.8, though GASfM receives ground-truth intrinsics and final bundle adjustment. For 292 bonsai views, its solver takes 56 seconds versus COLMAP’s 988, after matching.

Summary by alphaXiv.

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