Rectifying Geometry-Induced Similarity Distortions for Real-World Aerial–Ground Person Re-Identification

作者
Kailash Hambarde,Hugo Pedro Martins Carriço Proença
出处
期刊:IEEE Transactions on Information Forensics and Security [Institute of Electrical and Electronics Engineers]
卷期号:: 1-1 被引量:1
标识
DOI:10.1109/tifs.2026.3744610
摘要

Aerial ground person re-identification (AG-ReID) is fundamentally challenged by extreme viewpoint and distance discrepancies between aerial and ground cameras, which induce severe geometric distortions and invalidate the assumption of a shared similarity space across views. The existing methods primarily use geometry aware feature learning or appearance-conditioned prompting, and implicitly assume that the geometry invariant dot-product similarity used within ”attention” remains reliable under large viewpoint and distance variations. The main hypothesis in this work is that this assumption does not hold: extreme camera geometry and scale variations systematically distort the query–key similarity space, and heavily compromise attention-based matching, even when feature representations are partially aligned. To address this issue, we introduce the notion ofGeometry-Induced Query–Key Transformation(GIQT), a lightweight and low-rank module that explicitly rectifies the similarity space by conditioning query-key interactions on camera geometry. Rather than modifying the feature content or the attention formulation itself, GIQT adapts the similarity computation to compensate for dominant, geometry-induced anisotropic distortions. Building on this local similarity rectification, we consider a geometry-conditioned prompt generation mechanism as an auxiliary component that provides global and view-adaptive representation priors, that are derived directly from camera geometry. Our experiments were carried out on four aerial-ground person re-identification benchmarks, where the proposed framework attains state-of-the-art Rank-1 accuracy, together with the best or second-best mAP, and improves robustness under extreme, corrupted, and previously unseen geometric configurations. In practice, this work gives a step towards operation in high-altitude and large viewing-angle regimes. Also, such contribution is provided at minimal computational overhead with respect to competing methods. Code: https://github.com/kailashhambarde/GeoReID.git.

科研通智能强力驱动
Strongly Powered by AbleSci AI
科研通是完全免费的文献互助平台,具备全网最快的应助速度,最高的求助完成率。 对每一个文献求助,科研通都将尽心尽力,给求助人一个满意的交代。
实时播报
Owen的应助被风-FBDD采纳,获得10
刚刚
Lytoo完成签到 ,获得积分10
刚刚
应万言完成签到,获得积分0
1秒前
2秒前
3秒前
彭于晏的应助被jiajia采纳,获得10
3秒前
在水一方的应助被开心德地采纳,获得10
3秒前
4秒前
hi完成签到,获得积分10
5秒前
kk完成签到,获得积分10
5秒前
欣喜安莲完成签到,获得积分10
5秒前
6秒前
敏感问晴完成签到,获得积分10
6秒前
XZJ发布了新的文献求助10
7秒前
miao完成签到 ,获得积分10
9秒前
10秒前
且寄流心完成签到,获得积分10
10秒前
了尘完成签到,获得积分10
10秒前
求助文献发布了新的文献求助10
10秒前
11秒前
子翱完成签到 ,获得积分10
11秒前
田様的应助被清平乐采纳,获得10
11秒前
叶子完成签到,获得积分10
11秒前
白简发布了新的文献求助10
11秒前
12秒前
852的应助被壮观的晓露采纳,获得10
12秒前
秋风的应助被wg采纳,获得10
12秒前
敏感问晴发布了新的文献求助10
12秒前
12秒前
李梦晓发布了新的文献求助10
12秒前
13秒前
徐sir完成签到 ,获得积分10
14秒前
14秒前
乐乐的应助被ll采纳,获得10
15秒前
坐标系完成签到 ,获得积分10
15秒前
15秒前
Jasper的应助被jonan采纳,获得10
16秒前
SoilMan发布了新的文献求助10
17秒前
健壮璎完成签到,获得积分10
17秒前
科研通AI6.2的应助被芳华如梦采纳,获得10
17秒前
高分求助中
(应助此贴封号)【重要!!请各用户(尤其是新用户)详细阅读】【科研通的精品贴汇总】 10000
拉伸高分子熔体中缠结行为的模拟和小角中子散射研究 1700
The World Atlas of Wine 9th edition 1000
Sex, Aging, and Death in a Medieval Medical Compendium: Trinity College Cambridge MS R.14.52, Its Texts, Language, and Scribe 500
高分子基功能界面水蒸发材料与应用研究 500
外场增强的水汽发电研究 500
基于结构调控的功能化水诱导发电材料与器件研究 500
热门求助领域 (近24小时)
化学 材料科学 医学 生物 纳米技术 计算机科学 工程类 有机化学 化学工程 生物化学 物理 复合材料 内科学 催化作用 光电子学 人工智能 细胞生物学 无机化学 物理化学 数学
热门帖子
关注 科研通微信公众号,转发送积分 8188785
求助须知:如何正确求助?哪些是违规求助? 9576421
关于积分的说明 21847553
捐赠科研通 7786814
什么是DOI,文献DOI怎么找? 3435417
关于科研通互助平台的介绍 2559647
邀请新用户注册赠送积分活动 2482366