# AnimalCLEF26 @ CVPR & CLEF — Data

来源：https://www.kaggle.com/competitions/animal-clef-2026/data

采集时间：2026-10-19T07:18:31Z

The objective of this competition is to develop a model capable of discovering (clustering) individual animals from images. For feature extraction, we provide a pre-trained model [MegaDescriptor](https://huggingface.co/BVRA/MegaDescriptor-L-384), and for pre-training, we provide a large-scale dataset [WildlifeReID-10k](https://www.kaggle.com/datasets/wildlifedatasets/wildlifereid-10k) with 10k identities and around 140k images. Participants may also use any re-identification dataset from the [WildlifeDatasets](https://github.com/WildlifeDatasets/wildlife-datasets) package.

For testing, we have curated a dataset consisting of data of four species: (i) 🐆 Eurasian lynxes (Czech Republic), (ii) 🦎 salamanders (Czech Republic), (iii)🐢 loggerhead sea turtles (Zakynthos, Greece), and (iv) 🦎 Texas horned lizards (Texas, USA). The first three datasets are the same as in the previous [AnimalCLEF edition](https://www.kaggle.com/competitions/animal-clef-2025/data). The last dataset is new.

**The test dataset is split into two main subsets:**
- **📂 Train** – Contains labelled data about the individuals. It is not always present, which mimics the intended behaviour to cluster a new database of images.
- **🔍 Test** – Consists of images for clustering, where the task is to create clusters, each corresponding to one individual animal.

## File Structure

The dataset is organised into structured directories to allow easy access and processing. It consists of labelled images for known individuals (*Train*) and unlabeled images (*Test*) that require clustering, metadata, and a sample submission file.  

- **📂 `images/{LynxID2025, SalamanderID2025, SeaTurtleID2022, TexasHornedLizards}/train/`** – Contains labelled images of known individuals, serving as the reference set.  
- **🔍 `images/{LynxID2025, SalamanderID2025, SeaTurtleID2022, TexasHornedLizards}/test/`** – Contains images to be clustered.
- **📄 `metadata.csv`** – Provides dataset metadata, including training-test splits and individual identities.  
- **📑 `sample_submission.csv`** – A sample submission file illustrating the required format for predictions.  

## Data loading and description

Please check the provided [notebook](https://www.kaggle.com/code/picekl/animalclef2026-starter-notebook). The simplest way of loading the dataset is via the [wildlife-datasets](https://github.com/WildlifeDatasets/wildlife-datasets) package via

```
from wildlife_datasets.datasets import AnimalCLEF2026

dataset = AnimalCLEF2026(root)
```

This is the dataset with metadata

```
dataset.metadata
```

containing the following columns:

*   `image_id` - id of the image. Only for internal purposes.
*   `identity` - identity of individual animals in the training set.
*   `path` - path of the image.
*   `date` - date when the image was taken (can be missing).
*   `orientation` - orientation of the relevant body part (can be missing).
*   `species` - which species the photo depicts.
*   `split` - whether it is the train or the test set.
*   `dataset` - which of the 2 datasets the image belongs to.

## Sample submission

The above-mentioned notebook provides a code on how to create a sample submission.