OpenAnimals: Revisiting Person Re-Identification for Animals Towards Better Generalization

TL;DR

ARBase is proposed, a strong model tailored for animal re-identification, which incorporates insights from extensive experiments and introduces simple yet effective animal-oriented designs, and demonstrates that ARBase consistently outperforms existing baselines, achieving state-of-the-art performance across various benchmarks.

Abstract

This paper addresses the challenge of animal re-identification, an emerging field that shares similarities with person re-identification but presents unique complexities due to the diverse species, environments and poses. To facilitate research in this domain, we introduce OpenAnimals, a flexible and extensible codebase designed specifically for animal re-identification. We conduct a comprehensive study by revisiting several state-of-the-art person re-identification methods, including BoT, AGW, SBS, and MGN, and evaluate their effectiveness on animal re-identification benchmarks such as HyenaID, LeopardID, SeaTurtleID, and WhaleSharkID. Our findings reveal that while some techniques generalize well, many do not, underscoring the significant differences between the two tasks. To bridge this gap, we propose ARBase, a strong \textbf{Base} model tailored for \textbf{A}nimal \textbf{R}e-identification, which incorporates insights from extensive experiments and introduces simple yet effective animal-oriented designs. Experiments demonstrate that ARBase consistently outperforms existing baselines, achieving state-of-the-art performance across various benchmarks.

Figures (6)

• Figure 1: Illustration of various species.
• Figure 2: Illustration of main modules in OpenAnimals.
• Figure 3: Illustration of ARBase. IBN, IN and BN for Instance-Batch, Instance and Batch Normalization, $[N_{1}, N_{2}, N_{3}, N_{4}]$ for number of blocks in each stage (e.g., $[3, 4, 6, 3]$ for ResNet50), CA for Cosine Annealing, $L_{tp}$ and $L_{ce}$ for Triplet and Cross-Entropy Loss.
• Figure 4: Illustration of BNNeck in BoT luo2019bag. FC and BN for Fully-Connected Layer and Batch Normalization, $L_{tp}$ and $L_{ce}$ for Triplet and Cross-Entropy Loss.
• Figure 5: Illustration of Multi-Branch Architectures in MGN wang2018learning. HS for Horizontal Split.