MOO (Multi-view Oriented Observations) is a synthetic dataset for cattle re-identification. It comprises 128,000 images of 1,000 individuals. Designed for systematic viewpoint analysis in AG-ReID, each image is annotated with its azimuth and elevation, camera calibration data, and depth maps. MOO enables rigorous study of the impact of viewpoint variations on identification performance, and serves as an effective pre-training source for transfer to real-world data.
MOO is publicly available at https://github.com/TurtleSmoke/MOO, and on Hugging Face at CEAai/MOO.
Viewpoints
Each individual is rendered from 128 viewpoints at a fixed camera distance, with 16 azimuths over the full circle and 8 elevations from −20° to 85°.
Citation
@article{grolleau2026moo, author = {Grolleau, William and Chaouch, Achraf and Sabourin, Astrid and Lapouge, Guillaume and Achard, Catherine}, title = {{MOO}: {A} Multi-view Oriented Observations Dataset for Viewpoint Analysis in Cattle Re-Identification}, journal = {arXiv preprint arXiv:2603.04314}, note = {Presented at the CVPR 2026 Workshop on Computer Vision for Animal Behavior Tracking and Modeling (CV4Animals), non-archival}, year = {2026}, doi = {10.48550/arXiv.2603.04314},}