Reproduce “PGGT-Loc: Primitive-Grounded Geometry Transformer for Feed-Forward Camera Localization” 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

Hanqiao Ye, Yuzhou Liu, Mengqi Rong, Jian Liu, Yangdong Liu, Shuhan Shen

PGGT localizes cameras by comparing image patches with planar map primitives and predicting pose in one forward pass. On ScanNet, Ray-based PGGT reaches 69.0% pose recall at joint thresholds of 0.2 m and 10°, versus 48.5% for PlanaReLoc +Refine. A separate layout task reaches 95.5% 2D pose recall; the tasks are trained separately and tests focus on planar indoor maps.

Summary by alphaXiv.

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