This paper introduces UAVSat-Deg, a robustness benchmark for UAV-satellite cross-view geo-localization built from University-1652-Deg and SUES-200-Deg. It covers 27 corruption types, including 19 core and 8 compound corruptions, across three severity levels, both retrieval directions, and multiple UAV heights, with more than 11.7 million pre-generated corrupted test images. The proposed ReLATE framework estimates token-level reliability, aggregates trustworthy local evidence, and fuses regulated query, CLS-token, and GeM-pooled representations. The authors report the best average corrupted-test performance among compared methods while retaining competitive clean-image accuracy.
No heat snapshots are available in the last 24 hours.