ISPRS-Annals ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences ISPRS-Annals ISPRS Ann. Photogramm. Remote Sens. Spatial Inf. Sci. 2194-9050 Copernicus Publications Göttingen, Germany 10.5194/isprs-annals-XI-2-2026-455-2026 Zero-shot Vision-Language Reranking for Cross-View Geolocalization Erzurumlu Yunus Talha 1 Anderson John E. 2 Shuart William J. 2 Toth Charles https://orcid.org/0000-0001-9461-4887 3 Yilmaz Alper 3 Dept. of Electrical and Computer Engineering, The Ohio State University, 281 W Lane Ave, Columbus, Ohio, USA US Army Corps of Engineers Geospatial Research Lab, Corbin Field Station, Woodford, Virginia, USA Dept. of Civil Engineering, The Ohio State University, 281 W Lane Ave, Columbus, Ohio, USA 03 07 2026 XI-2-2026 455 461 Copyright: © 2026 Yunus Talha Erzurumlu et al. 2026 This work is licensed under the Creative Commons Attribution 4.0 International License. To view a copy of this licence, visit https://creativecommons.org/licenses/by/4.0/ This article is available from https://isprs-annals.copernicus.org/articles/XI-2-2026/455/2026/isprs-annals-XI-2-2026-455-2026.html The full text article is available as a PDF file from https://isprs-annals.copernicus.org/articles/XI-2-2026/455/2026/isprs-annals-XI-2-2026-455-2026.pdf

Cross-view geolocalization (CVGL) systems, while effective at retrieving a list of relevant candidates (high Recall@k), often fail to identify the single best match (low Top-1 accuracy). This work investigates the use of zero-shot Vision-Language Models (VLMs) as rerankers to address this gap. We propose a two-stage framework: state-of-the-art (SOTA) retrieval followed by VLM reranking. We systematically compare two strategies: (1)<strong> Pointwise</strong> (scoring candidates individually) and (2) <strong>Pairwise</strong> (comparing candidates relatively). Experiments on the VIGOR dataset show a clear divergence: all pointwise methods cause a catastrophic drop in performance or no change at all. In contrast, a <strong>pairwise comparison</strong> strategy using LLaVA<strong> improves Top-1 accuracy</strong> over the strong retrieval baseline. Our analysis concludes that, these VLMs are poorly calibrated for <em>absolute</em> relevance scoring but are effective at fine-grained <em>relative</em> visual judgment, making pairwise reranking a promising direction for enhancing CVGL precision.