Animal re-identification
Animal re-identification
Embedded Files
Animal re-identification is the process of recognizing individual animals within a population by using unique features such as markings, patterns, colors, or shapes. It is a vital tool in wildlife conservation and research, enabling precise tracking of animal movements, behavioral studies, or population monitoring without invasive measures like tagging. We provide a series of tools for the development of AI models, which significantly reduce the data processing time. These include animal re-identification datasets, pre-trained AI models or organizing competitions for raising awareness.
Python library providing an API to download and handle 50 animal re-identification datasets and 2 metadatasets.
Includes utilities to mass download and convert them into a unified format, and fix some wrong labels.
Received the Best Paper Award at WACV 2024.
An AI model trained on the datasets from the WildlifeDatasets library.
Widely used in the scientific community with 70k+ downloads.
It can be coupled with the other re-identification models using the WildFusion technique.
Novel datasets
We publish new datasets to help in training AI models.
SeaTurtleID2022: A novel dataset of loggerhead sea turtles of 7k+ images of 438 individuals.
WildlifeReID-10k: A compilation dataset of 140k+ images of 10k+ individuals. Due to the cleaning and unification, it aims to become the new benchmarking dataset for animal re-identification.
The first edition of the re-identification challenge attracted over 200 participants.
Included novel images of lynxes, sea turtles, and salamanders.
The participants managed to significantly beat the provided baselines, giving feedback to the scientific community for developing new methods.
Page updated
Google Sites
Report abuse