核心概念
提案されたSDPLは、位置のシフトやスケールの変化に対して堅牢であり、画像検索タスクで競争力のある性能を達成します。
要約
提案されたSDPLは、画像を複数の部分に分割し、コンテキスト情報を探索する密なパーティション戦略(DPS)と特徴シフト戦略を導入しています。これにより、位置のずれやスケールの変化に対処し、精度の高いクロスビュー地理位置情報を実現します。SDPLはUniversity-1652およびSUES-200という2つの主要なベンチマークデータセットで競争力のあるパフォーマンスを達成しました。この研究では、画像検索タスクにおいて新しいアプローチが提案され、その有効性が実証されました。
統計
Extensive experiments show that our SDPL achieves state-of-the-art performance on two prevailing benchmarks, i.e., University-1652 and SUES-200.
Existing methods mainly focus on digging more comprehensive information through feature maps segmentation, while inevitably destroy the image structure and are sensitive to the shifting and scale of the target in the query.
To address the above issues, we introduce a simple yet effective part-based representation learning, called shifting-dense partition learning (SDPL).
Our main contributions are as follows:
• We propose a shifting-dense partition learning (SDPL), including dense partition strategy and shifting-fusion strategy, to achieve accurate cross-view geo-localization against position deviations and scale changes.
• To mitigate the impact of scales changes, we propose the DPS to divide feature into shape-diverse parts, thus mining fine-grained representation while preserving global structure.
• For the degradation caused by position offset, we design the shifting-fusion strategy, which generates multiple sets of partitions with various segmentation centers, and then adaptively highlights the partitions that are consistent with target spatial distribution.
引用
"Existing methods mainly focus on digging more comprehensive information through feature maps segmentation."
"To address the above issues, we introduce a simple yet effective part-based representation learning, called shifting-dense partition learning (SDPL)."