LoRetta: A Foundation Model and Extensive Dataset for Global-Scale Remote Sensing Dense Image Matching
Quick Answer
LoRetta is a foundation model designed for global-scale remote sensing dense image matching, achieving an AUC of 83.3% on the LEVIR-GM benchmark.
Quick Take
It outperforms the baseline RoMa v2 by 1.6 points while reducing inference latency by 47.8%. The model integrates matchability-aware localization with guided dense registration, addressing challenges in varying image conditions.
Key Points
- LoRetta reformulates dense matching as localization-and-registration for improved accuracy.
- LEVIR-GM dataset includes 103K aligned and 827K augmented pairs across six continents.
- The model shows significant improvements in correct keypoints at 1 and 2 pixels.
- LoRetta reduces inference latency by 47.8%, enhancing operational efficiency.
- Geolocalization experiments confirm LoRetta's transferability across different platforms.
DeepSignal Analysis
What happened
LoRetta is a new foundation model for dense image matching in global-scale remote sensing. It achieves an AUC of 83.3% on the LEVIR-GM benchmark, surpassing the previous best model, RoMa v2, by 1.6 points while significantly reducing inference latency by 47.8%. The model addresses challenges posed by varying image conditions through a novel approach that combines localization and registration.
Key evidence
- LoRetta achieves an area under the curve (AUC) of 83.3% on the LEVIR-GM benchmark, outperforming RoMa v2 by 1.6 points.
- The model reduces inference latency by 47.8%, indicating improved efficiency in processing.
- LEVIR-GM is a global-scale multi-temporal optical matching benchmark consisting of 103K aligned and 827K augmented pairs across six continents over five years.
Why it matters
The development of LoRetta represents a significant advancement in remote sensing technology, particularly for applications requiring accurate dense image matching across diverse conditions. Its ability to efficiently handle large geometric offsets and varying image characteristics can enhance the reliability of geospatial analyses. This could have implications for fields such as environmental monitoring, urban planning, and disaster response, where timely and accurate data is crucial.
Paper Resources
📖 Reader Mode
~2 min readAbstract:Dense image matching establishes pixel-wise correspondences and underpins broad applications in computer vision and photogrammetry. However, extending dense matching to global-scale remote sensing remains challenging because image pairs may differ in acquisition time, season, viewpoint, spatial resolution, and land-cover state. The resulting large geometric offsets, partial overlap, and intrinsically unmatchable regions make direct dense correspondence prediction unreliable and inefficient. We thus reformulate dense matching as localization-and-registration: first localizing the matchable overlap and affine geometry, then refining dense residuals within the aligned frame. Based on this formulation, we propose LoRetta, a foundation model coupling matchability-aware affine localization with guided dense registration. We also introduce LEVIR-GM, a global-scale multi-temporal optical matching benchmark with dataset-native matchability labels (103K aligned, 827K augmented pairs, six continents, five years, 0.5-1024 m resolution). We further establish a unified evaluation protocol for sparse, semi-dense, and dense matchers. On LEVIR-GM, LoRetta achieves an area under the curve (AUC) of 83.3%, outperforming the strongest baseline RoMa v2 by 1.6 points, with larger percentage of correct keypoints (PCK) gains of 6.5 and 8.2 points at 1 and 2 pixels, while reducing inference latency by 47.8%. Astronaut-to-satellite and unmanned aerial vehicle (UAV)-to-satellite geolocalization experiments further demonstrate its transferability as a reusable geometric aligner.
| Comments: | 17 pages, 12 figures, 6 tables. Submitted to IEEE Transactions on Pattern Analysis and Machine Intelligence. Project page: this https URL |
| Subjects: | Computer Vision and Pattern Recognition (cs.CV); Image and Video Processing (eess.IV) |
| Cite as: | arXiv:2608.04106 [cs.CV] |
| (or arXiv:2608.04106v1 [cs.CV] for this version) | |
| https://doi.org/10.48550/arXiv.2608.04106 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Siwei Yu [view email]
[v1]
Tue, 4 Aug 2026 18:02:19 UTC (40,318 KB)
— Originally published at arxiv.org
Want this in your inbox every morning?
Daily brief at your local 8am — bilingual EN/中文, free.
More from arXiv cs.CV
See more →ProMoE-FL: Prototype-conditioned Mixture of Experts for Multimodal Federated Learning with Missing Modalities
ProMoE-FL introduces a Prototype-conditioned Mixture-of-Experts framework for multimodal federated learning, effectively addressing missing modalities. It outperforms existing methods on four chest X-ray datasets, demonstrating superior feature synthesis capabilities in both homogeneous and heterogeneous settings.