Multi‐scale attention encoder for street‐to‐aerial image geo‐localization

Abstract The goal of street‐to‐aerial cross‐view image geo‐localization is to determine the location query street‐view by retrieving aerial‐view from same place. drastic viewpoint and appearance gap between images brings a huge challenge against this task. In paper, we propose novel multiscale attention encoder capture contextual information aerial/street‐view images. To bridge domain these two view images, first use an inverse polar transform make approximately aligned with Then, explored applied convert into feature representation guidance learnt information. Finally, global mining strategy enable network pay more hard negative exemplars. Experiments on standard benchmark datasets show that our approach obtains 81.39% top‐1 recall rate CVUSA dataset 71.52% CVACT dataset, achieving state‐of‐the‐art performance outperforming most existing methods significantly.