OffNadirLoc: Benchmark and Framework for Challenging UAV-to-Satellite Geo-Localization under Large Off-Nadir Views

2026-07-23 · Source: cs.CV updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems · Depth: Expert, extended

Summary

The OffNadirLoc benchmark and ONLoc framework address challenging UAV-to-satellite geo-localization under large off-nadir views, where severe perspective distortions and occlusions occur. OffNadirLoc comprises 9,736 UAV images and 1,657 satellite images across 44 diverse regions, simulating extreme viewing angles (70°, 75°, 80°, 82°, 85°) and 1000x600 pixel resolution. The ONLoc framework, built on a DINOv2-Base encoder with 86M parameters, introduces a structure-aware contextual weighting mechanism to emphasize reliable local features and a view-coherent learning strategy for viewpoint-invariant features. Experiments show ONLoc outperforms leading approaches, achieving 6.9% higher Recall@1 and 4.75% higher AP on the OffNadirLoc benchmark, and demonstrates strong zero-shot generalization to near-nadir datasets.

Key takeaway

For Machine Learning Engineers developing UAV geo-localization systems, this work highlights the critical need to address large off-nadir viewing angles. You should integrate structure-aware contextual weighting to manage perspective distortions and adopt view-coherent learning strategies to leverage multi-view data. This approach significantly improves robustness and generalization, enabling more reliable deployments in complex real-world environments beyond near-nadir assumptions.

Key insights

Large off-nadir UAV-to-satellite geo-localization requires structure-aware feature weighting and multi-view coherent learning for robustness.

Principles

• Dynamically weight local features to suppress redundancy.
• Group multi-view UAV images with satellite for consistency.
• Prioritize geometrically reliable cues over repetitive textures.

Method

ONLoc extracts local and global features, projects them, computes redundancy scores for contextual weighting, and aggregates features into K=64 clusters. It then uses a view-coherent learning strategy with Multi-Similarity loss for group-wise supervision.

In practice

• Use DINOv2-Base as a robust image encoder.
• Implement redundancy-aware clustering for distorted imagery.
• Apply group-level supervision for multi-view consistency.

Topics

• UAV Geo-localization
• Off-Nadir Imagery
• Cross-View Matching
• Structure-Aware Learning
• View-Coherent Learning
• DINOv2

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Computer Vision Engineer

Related on AIssential

• OffNadirLoc: Benchmark and Framework for Challenging UAV-to-Satellite Geo-Localization under Large Off-Nadir Views — OffNadirLoc and ONLoc advance UAV-to-satellite geo-localization by addressing large off-nadir views with structure-aware and view-coherent learning.
• RIM: A Retrieval-In-Matching Framework for Cross-Domain Global Visual Localization of UAVs — Retrieval-In-Matching (RIM) efficiently localizes UAVs by sharing a single foundation model backbone for both global retrieval and local matching.
• Visual Relocalization from Sparse Views in Aliased and Low-Texture Environments via Novel View Synthesis — Combining MVS and LiDAR depths for geometry-aware 3DGS training enables robust visual relocalization in challenging, low-texture environments.
• Leveraging Large Vision Model for Multi-UAV Co-perception in Low-Altitude Wireless Networks — A novel framework optimizes multi-UAV perception by sparsifying visual data and using LVMs with DRL for efficient communication.
• GAP-GDRNet: Geometry-aware monocular 6D pose estimation for spacecraft using synthetic geometric supervision — GAP-GDRNet enhances monocular 6D spacecraft pose estimation by integrating attention-based feature refinement and patch-level self-attention into the GDR-Net pipeline.
• Compact Object-Level Representations with Open-Vocabulary Understanding for Indoor Visual Relocalization — OpenReLoc uses object-level representations and open-vocabulary semantics for robust indoor visual relocalization.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by cs.CV updates on arXiv.org.