AnimalCLEF 2025

Schedule

December 2024: Registration opens for all LifeCLEF challenges Registration is free of charge
28 February 2025: Competition Start
12 May 2025: Competition Deadline
31 May 2025: Deadline for submission of working note papers by participants [CEUR-WS proceedings]
23 June 2025: Notification of acceptance of working note papers [CEUR-WS proceedings]
30 June 2025: Camera-ready deadline for working note papers.
9-12 Sept 2025: CLEF 2025 Madrid - Spain

All deadlines are at 11:59 PM CET on a corresponding day unless otherwise noted. The competition organizers reserve the right to update the contest timeline if they deem it necessary.

Motivation

Animal re-identification is crucial for studying various aspects of wildlife, including tracking populations, monitoring movements, examining behaviors, and managing species. Despite variations in how animal re-identification is defined and approached across studies, its objective remains consistent. The primary aim is to accurately and efficiently identify individual animals within a species by recognizing unique traits, such as markings, patterns, or other distinctive features. Automating this process allows for collecting detailed, large-scale data on population dynamics, migration routes, habitat use, and behaviors. This data empowers researchers to track movements, assess population sizes, and observe demographic changes, providing invaluable insights into species dynamics. Such information is essential for identifying biodiversity threats and crafting evidence-based conservation strategies. Despite great progress in machine learning, many models for animal re-identification still struggle with overfitting. They often focus too much on background details -- like lighting or landscape -- rather than the animal's unique features. As a result, these models perform well in familiar environments but fail in new or varied conditions. This limitation makes it harder to accurately identify animals in diverse habitats, reducing the models' usefulness in real-world conservation efforts. Improving their ability to generalize is essential for more reliable animal identification.

Task Description

This challenge is all about individual animal identification of the following three species: (i) loggerhead sea turtles (Zakynthos, Greece), (ii) salamanders (Czech Republic), and (iii) Eurasian lynxes (Czech Republic). Your goal will be to design a model that, for each image, determines whether the depicted animal is new (not present in the training dataset) or known (in which case, its identity must be provided).

Participants may decide to work with the relatively small provided dataset or to boost the model performance by employing the WildlifeReID-10k: dataset: a collection of 36 existing wildlife re-identification datasets, with additional processing and diverse species such as marine turtles, primates, birds, African herbivores, marine mammals, and domestic animals. WildlifeReID-10k contains approximately 140,000 images of over 10,000 individuals.

Participation requirements

1. Subscribe to CLEF (AnimalCLEF task) by filling this form
2. Submit your solution before the competition deadline. More info on a Kaggle competition platform.

Publication Track

All registered participants are encouraged to submit a working-note paper to peer-reviewed LifeCLEF proceedings (CEUR-WS) after the competition ends.
This paper must provide sufficient information to reproduce the final submitted runs.

Only participants who submitted a working-note paper will be part of the officially published ranking used for scientific communication.

The results of the campaign appear in the working notes proceedings published by CEUR Workshop Proceedings (CEUR-WS.org).
Selected contributions among the participants will be invited for publication in the Springer Lecture Notes in Computer Science (LNCS) the following year.

Data

The objective of this competition is to develop a model capable of identifying individual animals from images. For pre-training, we provide a new large-scale dataset WildlifeReID-10k with 10k identities and around 140k images.

For testing, we have curated a dataset consisting of never-seen data of three species: (i) loggerhead sea turtles (Zakynthos, Greece), (ii) salamanders (Czech Republic), and (iii) Eurasian lynxes (Czech Republic). The test dataset is split into two main subsets:

Database – Contains labeled data about the individuals.
Query – Consists of images for identification, where the task is to determine whether the individual is known (present in the database) or new (not present in the database).

More info on a Kaggle competition platform.

Evaluation process

There are two metrics computed:

BAKS (balanced accuracy on known samples) - computes the accuracy for individuals which are known (in the database). It is balanced over classes (individuals).
BAUS (balanced accuracy on unknown samples) - computes the accuracy for individuals which are unknown (not in the database). It is balanced over classes (individuals).

The final accuracy is the geometric mean of the two accuracies. We decided to use the geometric mean instead of the arithmetic mean because the submission predicting all images as new individuals achieves 0% BAKS but 100% BAUS and its arithmetic mean would be 50% even though the model is useless. We provide a full implementation of the evaluation metric.