CzechLynx Dataset (v1.0)

The CzechLynx dataset includes real camera trap photographs and synthetic samples of the Eurasian lynx (Lynx lynx), organized around three downstream tasks:

1. individual identification, 
2. pose estimation, and
3. instance segmentation. 

The main part of the dataset, consisting of 39,760 manually verified and labeled camera-trap images, is fixed, whereas the synthetic part, in practice, can be scaled to any size (for simple use, a synthetic subset with a similar number of individuals and images is provided above in the CzechLynx.zip file. The real images span more than 15 years and come from two geographically distinct regions in Central Europe: Southwest Bohemia and the Western Carpathians.

All images are stored in JPEG format (with 90% compression), with metadata provided in a structured CSV file.
To simplify access to the data and support standardized development and evaluation of downstream tasks, the Images  and Metadata are distributed in a single zip file, even though not all components are required for every task. Instead of maintaining separate annotation files for each downstream task, a single shared CSV file with all annotations and necessary information is provided.

Summary of data sources

Metadata

Most images in the CzechLynx dataset come with rich metadata that help you understand when, where, and who was captured, and how each image can be used in downstream tasks. For each observation, we include basic provenance (which monitoring project it came from), temporal information (date of capture, how long since the individual was first seen, and an encounter ID for sequences from the same trap), and spatial context (10 × 10 km ETRS89-LAEA grid-cell code, nearest administrative region, trap ID, and centroid GPS coordinates).

On top of that, we provide phenotypic labels (lynx coat pattern), computer-vision annotations (instance segmentation masks and 2D pose keypoints), and flags for predefined dataset splits (geo-aware, time-aware open/closed, and pose splits). Together, these fields make it easy to filter and group images by identity, time, space, appearance, or benchmark split, so you can quickly set up reproducible experiments for re-identification, pose estimation, and segmentation.

Task-specific subsets

The CzechLynx dataset is organized into three subsets tailored for: (i) Individual re-identification, (ii) Animal pose estimation, and (iii) Instance segmentation. Each subset largely overlaps but differs by inclusion criteria and annotation detail. For individual identification and instance segmentation, the same set
of images with clearly visible coat patterns, for which human experts confirm the identity, is provided. Each image is paired with an identity label and a pixel-level mask outlining the lynx body, which enables the training of segmentation models while providing suitable input for re-identification. The pose estimation part is a subset of the identification/segmentation images and is smaller due to the labor-intensive annotation process.

Predefined Splits

To support robust evaluation under real-world constraints, CzechLynx provides three distinct splits:

• Geo-aware-open: Train on Carpathians, test on southwest Bohemia (disjoint individuals).
• Time-aware-open: Train on the earlier period, test on the later period with some unseen individuals.
• Time-aware-closed: Train/test split by time, all identities in test appear in training.