Search papers, labs, and topics across Lattice.
The authors introduce MOO, a large-scale synthetic dataset for animal re-identification containing 128,000 images of 1,000 cattle individuals across 128 viewpoints, designed to analyze the impact of viewpoint variations. Using MOO, they quantify the influence of elevation angle on ReID performance, identifying a critical elevation threshold for improved generalization. Experiments demonstrate the transferability of insights from MOO to real-world cattle ReID datasets, showing performance gains in both zero-shot and supervised settings.
Synthetic data reveals a critical elevation threshold in aerial-ground animal re-identification, above which models generalize significantly better to unseen viewpoints.
Animal re-identification (ReID) faces critical challenges due to viewpoint variations, particularly in Aerial-Ground (AG-ReID) settings where models must match individuals across drastic elevation changes. However, existing datasets lack the precise angular annotations required to systematically analyze these geometric variations. To address this, we introduce the Multi-view Oriented Observation (MOO) dataset, a large-scale synthetic AG-ReID dataset of $1,000$ cattle individuals captured from $128$ uniformly sampled viewpoints ($128,000$ annotated images). Using this controlled dataset, we quantify the influence of elevation and identify a critical elevation threshold, above which models generalize significantly better to unseen views. Finally, we validate the transferability to real-world applications in both zero-shot and supervised settings, demonstrating performance gains across four real-world cattle datasets and confirming that synthetic geometric priors effectively bridge the domain gap. Collectively, this dataset and analysis lay the foundation for future model development in cross-view animal ReID. MOO is publicly available at https://github.com/TurtleSmoke/MOO.
