Emergent Extreme-View Geometry in 3D Foundation Models
This addresses the problem of robust 3D vision under challenging viewpoints for researchers and practitioners in computer vision, representing an incremental improvement through targeted adaptation.
The paper investigates 3D foundation models' ability to handle extreme, non-overlapping views and finds they exhibit emergent geometry understanding without specific training, then introduces a lightweight alignment scheme that improves relative pose estimation by 15-20% without degrading other outputs, while also releasing a new benchmark dataset.
3D foundation models (3DFMs) have recently transformed 3D vision, enabling joint prediction of depths, poses, and point maps directly from images. Yet their ability to reason under extreme, non-overlapping views remains largely unexplored. In this work, we study their internal representations and find that 3DFMs exhibit an emergent understanding of extreme-view geometry, despite never being trained for such conditions. To further enhance these capabilities, we introduce a lightweight alignment scheme that refines their internal 3D representation by tuning only a small subset of backbone bias terms, leaving all decoder heads frozen. This targeted adaptation substantially improves relative pose estimation under extreme viewpoints without degrading per-image depth or point quality. Additionally, we contribute MegaUnScene, a new benchmark of Internet scenes unseen by existing 3DFMs, with dedicated test splits for both relative pose estimation and dense 3D reconstruction. All code and data will be released.