Reproduce “FlexGlue: Sparse-to-Dense Feature Matching and Visual Disambiguation” 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

Philipp Lindenberger, Paul-Edouard Sarlin, Ondrej Miksik, Marc Pollefeys

FlexGlue combines sparse point matching with pair-level overlap prediction to reject look-alike images during 3D mapping. It contextualizes dense image features, then reads and refines matches at selected points, with early exit for confident non-overlap. On 25-scene ETH3D reconstruction, SuperPoint with FlexGlue reports a 5° pose AUC of 92.3; this end-to-end comparison also changes the matcher and filter.

Summary by alphaXiv.

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