A Cross-View Matching Method Based on Dense Partition Strategy for UAV Geo-localization

What is it about?

This paper reports our solution for ACM Multimedia 2023 cross-view geo-localization challenge, which aims to solve real-world geo-localization task with extremely large satellite-view gallery distractors. Our solution is built on the basis of SwinV2 and LPN. Concretely, we adopt the SwinV2-B, the current mainstream transformer-based feature extractor, as the backbone of our model. Inspired by the feature partition strategy of LPN, we design a more efficient partition strategy named dense partition strategy. It segments and combines features to alleviate the problem of feature discontinuity at the boundary of different partitions. We get eight place in the geographic localization track on the official leaderboard

Featured Image

Photo by Humphrey Muleba on Unsplash

Photo by Humphrey Muleba on Unsplash

Why is it important?

we design an end-to-end model, called Dense Partition Network (DPN), to solve the geo-location task from the perspective of drones. Our main techniques include: 1. We adopt an efficient feature extractor SwinV2 as backbone. The transformer-based structure has a strong feature representation capability, allowing the model to achieve satisfactory recall and accuracy without additional modules. 2. Inspired by LPN [22], we propose a more considerate partition strategy name dense partition strategy. We combine the basic blocks of LPN to obtain multi-scale and multi-type partition blocks, so as to effectively alleviate the performance degradation caused by the spatial position offset and scale uncertainty of the target.

Perspectives