MCL-Geo: Multi-branch Contrastive Learning for Cross-view Geo-localizationA multi-branch contrastive learning framework plus uniform cross-view contrastive loss for cross-view geo-localization targets

Cross-view geo-localization aims to match images of the same location taken from different platforms, such as drones and satellites. However, the altitude differences between the perspectives of these platforms can lead to irregularities in the appearance of the target object. To obtain cross-view invariant features, existing methods pair samples from different perspectives and use contrastive learning between analogous samples. However, these methods are influenced by the backbone, while overlooking more distinctive features. Therefore, in order to effectively learn the consistency information between images across views, we propose a multi-branch framework combine a uniform contrastive loss function which can simultaneously mine the consistency information between multiple samples. In addition, we propose a novelty special designed data processing strategy to process the input of contrastive learning model, which can prompt the model learning fine-grained invariant details about the target between different perspectives. Experiments on widely used public benchmarks show that our proposed method achieves superior performance with fewer parameter models.