Learning Cross-view Visual Geo-localization without Ground Truth

arXiv:2403.12702v1 Announce Type: new
Abstract: Cross-View Geo-Localization (CVGL) involves determining the geographical location of a query image by matching it with a corresponding GPS-tagged reference image. Current state-of-the-art methods predominantly rely on training models with labeled paired images, incurring substantial annotation costs and training burdens. In this study, we investigate the adaptation of frozen models for CVGL without requiring ground truth pair labels. We observe that training on unlabeled cross-view images presents significant challenges, including the need to establish relationships …

abstract annotation art arxiv costs cs.cv current geo gps image images localization location query reference state study training training models truth type view visual