Weather-Robust Cross-View Geo-Localization via Prototype-Based Semantic Part Discovery

TL;DR

This paper introduces SkyPart, a lightweight, prototype-based semantic part discovery method for cross-view geo-localization that improves robustness to weather and altitude variations, setting new state-of-the-art results.

Contribution

SkyPart is a novel patch grouping head for vision transformers that explicitly models semantic parts and incorporates altitude-conditioned modulation and uncertainty-weighted loss.

Findings

SkyPart achieves state-of-the-art accuracy on multiple datasets.

It is the smallest among top-performing methods with 26.95M parameters.

SkyPart maintains high performance under weather-related corruptions.

Abstract

Cross-view geo-localization (CVGL), which matches an oblique drone view to a geo-referenced satellite tile, has emerged as a key alternative for autonomous drone navigation when GNSS signals are jammed, spoofed, or unavailable. Despite strong recent progress, three limitations persist: (1) global-descriptor designs compress the patch grid into a single vector without separating layout from texture across the view gap; (2) altitude-related scale variation is retained in the learned embedding rather than marginalized; and (3) multi-objective training relies on hand-tuned scalars over losses on incompatible gradient scales. We propose SkyPart, a lightweight swappable head for patch-based vision transformers (ViTs) that institutes explicit part grouping over the patch grid. SkyPart has four theory-grounded components: (i) learnable prototypes competing for patch tokens via single-pass cosine assignment; (ii) altitude-conditioned linear modulation applied only during training, making the retrieval embedding altitude-free at inference; (iii) a graph-attention readout over active prototypes; and (iv) a Kendall uncertainty-weighted multi-objective loss whose stationary points are Pareto-stationary. At 26.95M parameters and 22.14 GFLOPs, SkyPart is the smallest among top-performing methods and sets a new state of the art on SUES-200, University-1652, and DenseUAV under a single-pass, no-re-ranking, no-TTA protocol. Its advantage over the strongest baseline widens under the ten-condition WeatherPrompt corruption benchmark.