Reproduce “Refin3R: Geometry-Guided Refinement for Feed-Forward 3D Reconstruction” 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

Jongmin Lee, Jaeyeong Kim, Seungryong Kim

Refin3R refines dense multi-view point maps by combining feed-forward predictions with sparse Structure-from-Motion (SfM) points and cameras. Its image-grid and 3D branches spread geometric guidance using image context and evidence from multiple views. On ETH3D with 16 views, it reaches 53.6% AUC@1 cm versus GGPT’s 32.6%. It depends on usable SfM guidance and does not guarantee recovery in uncovered regions.

Summary by alphaXiv.

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