> **_NOTE:_**  The competition is over. We would like to thank all participants for taking part and sharing their ideas. To promote further learning, we compiled the list of participants' [codes and working notes](https://www.kaggle.com/competitions/animal-clef-2026/discussion/721036). See you next year :)

This challenge focuses on **identifying individual animals** through image-based re-identification in a realistic, field-style setup. Instead of only matching images within a closed catalog, the goal combines **lookup + discovery**: use the provided **identity database** (for selected species) when possible, and **group the remaining test images into individuals** when they do not match any known identity. In other words, you must handle both **known** individuals and **new** individuals in the test set. The competition follows a **discovery setting** in the sense that many test individuals are **not present in training**, and the task includes assigning new images to previously unseen identities.

* **For training / database**, we provide images (with identities) for **three species**: 🐆 *Lynx lynx*, 🦎 *Salamandra salamandra*, and 🐢 *Caretta caretta*. Participants may also boost performance by leveraging the [WildlifeReID-10k](https://openaccess.thecvf.com/content/CVPR2025W/FGVC/papers/Adam_WildlifeReID-10k_Wildlife_re-identification_dataset_with_10k_individual_animals_CVPRW_2025_paper.pdf) dataset, available via the [WildlifeDatasets](https://github.com/WildlifeDatasets/wildlife-datasets) library.
* **The test set** consists of images from **four species**: 🐆 *Lynx lynx*, 🦎 *Salamandra salamandra*, 🐢 *Caretta caretta*, and 🐉 *Phrynosoma cornutum*. For the first three species, some test images may correspond to **known individuals** from the provided database, while others will be **new individuals** that must be discovered and grouped.

**Submissions will be evaluated using the [Adjusted Rand Index](https://en.wikipedia.org/wiki/Rand_index) (ARI)**, which measures agreement between predicted groupings and ground truth. ARI penalizes both **over-clustering** (splitting one individual into multiple groups) and **under-clustering** (merging different individuals into the same group).