Researchers have proposed a one-stage end-to-end model for fine-grained wildlife re-identification. The model uses DINOv2 for spatial geometry and MegaDescriptor for re-identification, and enhances latent queries with prompt features. Preliminary findings show that the model achieves a competitive mean average precision score of 30.584%, compared to a state-of-the-art two-stage approach which reaches 44.89%. Qualitative results demonstrate effective bounding and identification of animal identities. --- Why it matters: This matters because it presents an alternative approach for fine-grained wildlife re-identification, potentially improving the efficiency and accuracy of such tasks. The model's competitive performance suggests that one-stage end-to-end models could be a viable solution for this challenging problem. Source: https://arxiv.org/abs/2608.18246