英文标题：Cross-Species Animal Re-Identification with Semantic Consistency Learning
作者：Shuoyi Chen, Yuejia Li, Mang Ye
arXiv ID：2609.09705 | 分类：cs.CV | 发表：2026-09-09
许可：CC-BY-SA
摘要 可泛化的动物重识别（ReID）旨在跨具有多样形态和生态环境的物种识别动物个体。与人重识别不同——在人重识别中不同域共享相似的身体结构——动物物种往往表现出截然不同的解剖结构和视觉模式，使得建立共享的视觉对应关系变得困难。因此，跨物种学习到的表征往往形成碎片化的嵌入空间，严重限制了跨物种泛化能力。{{NL}}为应对这一挑战，我们提出了语义一致性学习（SCL）框架，旨在学习在外观变化下保持稳定、同时保留跨物种共享语义结构的表征。SCL由两个互补组件构成。前景–背景解耦谱归一化（FDSNorm）通过以区域感知的方式抑制环境引起的风格变化来稳定特征统计量，而跨物种邻域建模（CNM）则通过动态特征
Generalizable animal Re-Identification (ReID) aims to recognize individual animals across species with diverse morphologies and ecological contexts. Unlike person ReID, where different domains share similar body structures, animal species often exhibit drastically different anatomical structures and visual patterns, making it difficult to establish shared visual correspondences. As a result, representations learned across species tend to form fragmented embedding spaces, which severely limits cross-species generalization. To address this challenge, we propose Semantic Consistency Learning (SCL), a framework designed to learn representations that remain stable across appearance variations while preserving semantic structures shared across species. SCL consists of two complementary components. Foreground-Background Decoupled Spectral Normalization (FDSNorm) stabilizes feature statistics by suppressing environment-induced style variations in a region-aware manner, while Cross-species Neighborhood Modeling (CNM) captures transferable relational structures across species through dynamic feature neighborhoods. Extensive experiments on 11 public animal ReID datasets demonstrate that SCL consistently outperforms state-of-the-art methods under multiple cross-species evaluation protocols and generalizes effectively to previously unseen species and ecological domains. Code is available at https://github.com/Kemalau/ECCV-26-SCL.
查看完整双语翻译 →
正在跳转到翻译阅读页… 如果没有自动跳转，请点击这里。