Baikal Rare Species: Nature and Environment Annotated Wildlife Dataset

Baikal Rare Species: Nature and Environment Annotated Wildlife Dataset is an annotated camera-trap image dataset collected in the Baikal region under real natural environmental conditions. The dataset is designed to support research in wildlife monitoring, biodiversity assessment, ecological image analysis, and machine learning methods for automated species recognition. The dataset contains camera-trap images captured across different seasons, weather conditions, illumination regimes, and landscape backgrounds. It includes both daytime RGB images and nighttime infrared images, reflecting the visual complexity of real-world protected-area monitoring. The data represent challenging natural scenes with variable object scale, partial animal visibility, occlusions, motion blur, low-contrast infrared imagery, dense vegetation, snow-covered backgrounds, and fixed camera locations that may introduce strong background correlations. The dataset includes annotated images of wildlife species characteristic of the Baikal region, including badger, bear, hare, wild boar, deer, wolf, and lynx. Several classes correspond to rare or underrepresented species, making the dataset particularly relevant for studying long-tailed class distributions, class imbalance, and model robustness on rare categories. The dataset is suitable for developing and evaluating computer vision models for wildlife image classification, animal detection, ecological monitoring, and domain adaptation under natural environmental variability. This dataset is intended to facilitate reproducible research on automated wildlife recognition in northern and taiga ecosystems. It may be used for benchmarking machine learning models under realistic camera-trap conditions, including seasonal domain shift, infrared night imaging, small or partially visible animals, and imbalanced species distributions. Potential applications Automated wildlife species classification from camera-trap images Biodiversity monitoring and ecological data analysis Rare species recognition under limited data conditions Evaluation of machine learning models under class imbalance Domain adaptation and robustness testing across seasons, lighting conditions, and camera locations Development of computer vision pipelines for protected-area monitoring The dataset is organized as compressed archives containing camera-trap images, while all annotations and metadata are provided in the accompanying dataset.json file. The JSON file stores image-to-label mappings and dataset annotation information required for machine learning and computer vision applications.