ReLATE paper adds 11.7M-image degraded UAV-satellite geo-localization benchmark

A team including Haochen Jiang, Jialei Pan, Yuzhe Sun, Zhe Dong, Lecheng Ren, Yanfeng Gu, and Tianzhu Liu propose ReLATE, a reliability-guided evidence fusion method for UAV-satellite cross-view geo-localization under real-world image degradations, in arXiv:2607.25524. UAV-satellite matching has achieved high accuracy on clean image benchmarks, the authors write, but real flights face adverse weather, illumination changes, platform motion, sensor noise, and compression. They introduce UAVSat-Deg, a large-scale robustness benchmark comprising University-1652-Deg and SUES-200-Deg. UAVSat-Deg covers 27 corruption types, including 19 core and 8 compound corruptions, at three severity levels. It supports bidirectional drone-to-satellite and satellite-to-drone retrieval and multi-height UAV acquisition, and contains more than 11.7 million pre-generated corrupted test images. Benchmarking representative methods showed substantial robustness gaps under severe and compound corruptions. ReLATE estimates a structure-smoothed reliability field over visual tokens, aggregates trustworthy local evidence, and combines regulated query representations with CLS-token and GeM-pooled branches into a final cross-view descriptor. Across both test sets and retrieval directions, the authors report that ReLATE achieved the best average corrupted-test performance among compared methods while remaining competitive on clean images. Code and dataset availability are noted in the paper.