Predictive Information Preservation via Variational Information Bottleneck for Cross-View Geo-Localization

Cross-view geo-localization task, which is to handle the problem of matching two images captured same target building, but from different viewpoints, e.g., satellite-view and drone-view, has received significant attention in recent years. However, this research is impeded by the large visual appearance changes across different views and irrelevant content contained in the background. Previous work mitigates the geo-view gap by some similarity-based constraints or utilizing rich contextual information near the target as auxiliary information. Despite some promising breakthroughs made by such methods, they fail to consider the involvement of irrelevant features retained in the high-dimensional features, which reduces the accuracy of the retrieval result. This paper proposes a simple and efficient model termed Predictive Information Preservation Bottleneck (PIPB), using the variational information bottleneck to discard the irrelevant information and retain the predictive information, enhancing the result performance. In particular, our proposed PIPB consists of two stages. Firstly, we learn the part-based features of each image to make full use of neighbor clues, which is realized by the square-ring partition strategy. Then, at the second stage, these learned representations are fed through the variational information bottleneck module to filter out superfluous information. This step can promote the robustness and generalization of our model and improve experiment performance. Extensive experiments are conducted on the recently-released dataset University-1652 and the fundamental benchmark CVACT, showing remarkable performance results compared to other competitive methods.