Abstract
We introduce a geometry-guided framework that refines dense feed-forward reconstructions using accurate geometric guidance obtained from an improved SfM pipeline, directly in 3D space. First, we revisit sparse-view SfM and introduce an improved pipeline that integrates dense matchers with a lightweight optimisation procedure. Second, we introduce a variant of a lightweight 3D Point Transformer that jointly processes dense point maps from feed-forward models and the geometrically grounded partial point cloud from our SfM pipeline, predicting residual corrections for every dense point.
