Zero-shot Vision-Language Reranking for Cross-View Geolocalization

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Erzurumlu, Yunus Talha, Anderson, John E., Shuart, William J., Toth, Charles, Yilmaz, Alper
Format: Preprint
Published: 2026
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918414481620992
author Erzurumlu, Yunus Talha
Anderson, John E.
Shuart, William J.
Toth, Charles
Yilmaz, Alper
author_facet Erzurumlu, Yunus Talha
Anderson, John E.
Shuart, William J.
Toth, Charles
Yilmaz, Alper
contents 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) Pointwise (scoring candidates individually) and (2) Pairwise (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 pairwise comparison strategy using LLaVA improves Top-1 accuracy over the strong retrieval baseline. Our analysis concludes that, these VLMs are poorly calibrated for absolute relevance scoring but are effective at fine-grained relative visual judgment, making pairwise reranking a promising direction for enhancing CVGL precision.
format Preprint
id arxiv_https___arxiv_org_abs_2603_27251
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Zero-shot Vision-Language Reranking for Cross-View Geolocalization
Erzurumlu, Yunus Talha
Anderson, John E.
Shuart, William J.
Toth, Charles
Yilmaz, Alper
Computer Vision and Pattern Recognition
Artificial Intelligence
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) Pointwise (scoring candidates individually) and (2) Pairwise (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 pairwise comparison strategy using LLaVA improves Top-1 accuracy over the strong retrieval baseline. Our analysis concludes that, these VLMs are poorly calibrated for absolute relevance scoring but are effective at fine-grained relative visual judgment, making pairwise reranking a promising direction for enhancing CVGL precision.
title Zero-shot Vision-Language Reranking for Cross-View Geolocalization
topic Computer Vision and Pattern Recognition
Artificial Intelligence
url https://arxiv.org/abs/2603.27251