Geometric Consistency Protocol for Foundation Model Features in Multi-View Satellite Imagery

The article introduces a Geometric Consistency Protocol designed for evaluating foundation model features in multi-view satellite imagery, addressing the inadequacies of conventional 2D global matching methods. It utilizes a Rational Function Model (RFM) framework to create a geometry-faithful and reproducible benchmark that integrates a 3D consistency metric with geometry-constrained matching. This work highlights the importance of geometric constraints in satellite imagery evaluation and reveals that state-of-the-art 2D models can still perform competitively against specialized 3D models under the new RPC-consistent evaluation criteria, which is crucial for practitioners focusing on accurate multi-view reconstruction in remote sensing applications.