CzechLynx

CzechLynx是一个大规模的公开数据集，用于对欧亚猞猁进行个体识别、2D姿态估计和实例分割。数据集包含超过3万张相机陷阱图像，这些图像被标注了分割掩码、身份标签和20点骨骼信息，并覆盖了15年中在两个地理上不同的区域（西南波希米亚和西喀尔巴阡山脉）的219个独特个体。为了增加数据的多样性，我们创建了一个互补的合成数据集，其中包含超过10万张通过Unity-based流程和扩散驱动的文本到纹理建模生成的逼真图像，涵盖了不同的环境、姿态和皮毛图案变化。为了允许跨空间和时间的泛化测试，我们定义了三个定制的评估协议/分割：(i) 地理感知，(ii) 时间感知开放集，和(iii) 时间感知封闭集。这个数据集旨在成为基准现有模型和发展新型方法的工具，不仅限于个体动物重新识别。

CzechLynx is a large-scale public dataset for individual identification, 2D pose estimation, and instance segmentation of Eurasian lynx. The dataset contains over 30,000 camera trap images, which are annotated with segmentation masks, identity labels, and 20-point skeletal information, covering 219 unique individuals across 15 years in two geographically distinct regions: Southwest Bohemia and the Western Carpathian Mountains. To enhance data diversity, we developed a complementary synthetic dataset containing over 100,000 photorealistic images generated via Unity-based pipelines and diffusion-driven text-to-texture modeling, which cover diverse environments, poses, and fur pattern variations. To enable generalization testing across space and time, we define three custom evaluation protocols/splits: (i) geography-aware, (ii) time-aware open-set, and (iii) time-aware closed-set. This dataset is intended to serve as a benchmark tool for evaluating existing models and developing novel methods, not limited to individual animal re-identification.