Zero-shot Vision-Language Reranking for Cross-View Geolocalization
For CVGL systems, this work identifies a practical reranking strategy to improve precision, though the improvement is incremental and dataset-specific.
Cross-view geolocalization suffers from low Top-1 accuracy despite high Recall@k. This work proposes a two-stage framework using zero-shot VLMs as rerankers, finding that pairwise comparison with LLaVA improves Top-1 accuracy over the retrieval baseline, while pointwise methods fail.
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.