Hyperbolic Visual Hierarchy Learning for Aerial-Ground Person Re-Identification
IEEE Signal Processing Letters · 2026 · 33 · 3093-3097
DOI: 10.1109/lsp.2026.3715427 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Signal Processing Letters, 2026. Verfasst von Xianxian Zeng und 5 weiteren Autor:innen. Die Arbeit nennt 29 Literaturangaben. Gefördert von: National Natural Science Foundation of China, Basic and Applied Basic Research Foundation of Guangdong Province, Guangdong Provincial Department of Education Key Field Project on Artificial Intelligence.
Abstract (Originalsprache)
Aerial-Ground Person Re-Identification (AGPReID) aims to match images of the same identity captured from UAV and ground camera viewpoints, and suffers from more severe cross-view domain shifts than conventional person re-identification. Existing local methods often treat different regions in isolation, lacking explicit modeling of inter-region dependencies and their varying importance. This paper proposes a Hyperbolic Relational Enhancement Framework (HREF), which leverages the natural suitability of hyperbolic geometry for hierarchical data to explicitly model patch-level relational structures within an image in hyperbolic space. The framework comprises two core modules: the Patch Relation Encoder (PRE) maps local features onto a hyperbolic manifold and employs hyperbolic attention to explicitly capture non-uniform inter-region relations; the Hierarchical Information Fusion (HIF) module interacts global features with PRE outputs to adaptively select the most discriminative relational information and inject it into the global representation. Experiments on the CARGO, AG-ReID, and AG-ReID.v2 datasets indicate that our method outperforms recent approaches, validating the effectiveness of hyperbolic relational modeling.
Förderung
- National Natural Science Foundation of China
- Basic and Applied Basic Research Foundation of Guangdong Province
- Guangdong Provincial Department of Education Key Field Project on Artificial Intelligence
- Guangdong Key Research Institute of Humanities and Social Sciences at Universities
- Key Research Institute of Digital Economy and New Quality Productive Forces
Zitieren
APA
Zeng, X., Chen, Y., Yuan, J., Yin, M., Xu, W., & Liu, S. (2026). Hyperbolic Visual Hierarchy Learning for Aerial-Ground Person Re-Identification. IEEE Signal Processing Letters, 33, 3093-3097. https://doi.org/10.1109/lsp.2026.3715427
Vancouver
Zeng X, Chen Y, Yuan J, Yin M, Xu W, Liu S. Hyperbolic Visual Hierarchy Learning for Aerial-Ground Person Re-Identification. IEEE Signal Processing Letters. 2026;33:3093-3097. doi:10.1109/lsp.2026.3715427
BibTeX
@article{Zeng2026,
author = {Zeng, Xianxian and Chen, Yijun and Yuan, Jun and Yin, Ming and Xu, Weichao and Liu, Shun},
title = {Hyperbolic Visual Hierarchy Learning for Aerial-Ground Person Re-Identification},
journal = {IEEE Signal Processing Letters},
year = {2026},
volume = {33},
pages = {3093--3097},
doi = {10.1109/lsp.2026.3715427}
}
EUVIMED