Solv3R: 3D Looped Transformer as a Multi-view Solver

Researchers introduced Solv3R, a 3D looped transformer that acts as a multi-view solver for structure-from-motion reconstruction, according to the paper's abstract. The work contrasts with most geometric foundation models, which the authors say do not support the iterative refinement that lets classical structure-from-motion expand maps without recomputing already computed views or improve geometry as more observations arrive. The paper presents Solv3R as a learned alternative that preserves that iterative, incremental reconstruction behavior.

Classical structure-from-motion reconstruction proceeds iteratively, refining camera poses and scene geometry from a current estimate. This allows maps to easily expand without recomputing already computed views, or geometry estimates to improve as more observations are provided. Most geometric foun