From Species-Specific Models to Universal Re-ID: A Survey of Animal Re-Identification

• Survey of 41 animal Re-ID studies published between 2020 and 2025. • Unified taxonomy covering feature, metric, temporal, multimodal, and training regimes. • Comprehensive review of datasets, benchmarks, and habitat-based evaluation protocols. • Analysis of major challenges, including domain shift, temporal drift, and annotation scarcity. • Future directions toward universal, multimodal, and ecologically robust Re-ID systems. Automated animal re-identification (Re-ID) has become an essential tool for wildlife ecology, conservation management, and precision livestock farming. Recent progress in deep representation learning, transformer architectures, multimodal learning, and vision–language modeling has accelerated the development of scalable, non-invasive systems for identifying individuals across images and videos. This survey provides a comprehensive review of animal Re-ID research published between 2020 and 2025, encompassing 41 peer-reviewed works. We propose a structured taxonomy of animal Re-ID methods and provide an integrated analysis of approaches, datasets, and evaluation practices. We also highlight persistent challenges, including domain shift, temporal variability, annotation scarcity, and inconsistent evaluation protocols, and outline broad future research directions toward universal, temporally robust, and ecologically meaningful animal Re-ID systems. This survey provides a unified foundation for advancing robust and deployable solutions in the coming decade.