EverWatch airborne bird imagery dataset for detection and classification of Everglades wading birds

Large-scale monitoring of populations and communities using airborne remote sensing is increasingly central to a broad range of research, management, and decision-making efforts. Performing this work at scale requires automating the detection and classification of individual organisms in airborne imagery. This is typically done using computer vision models, which require large amounts of data for training and precisely labeled data for evaluation. One area of high demand for airborne detection and classification is wildlife monitoring, including the monitoring of large birds. However, open mixed-species annotated datasets that are sufficiently large for training computer vision models are rare, especially for the United States. Mixed-species datasets are important because many bird species forage, roost, and nest in mixed-species flocks, requiring computer vision models for airborne monitoring to be able to accurately distinguish between similar-looking species from above. Here we provide a dataset of airborne imagery from uncrewed aircraft systems and associated detection and species classification labels of over 50,000 wading birds in the Florida Everglades. The dataset includes 5128 images with 50,491 annotations for training (split into training and validation data) and 196 images with 4113 annotations for model evaluation, making it the largest open mixed-species dataset of which we are aware. Species represented in the dataset include White Ibis (Eudocimus albus), Great Egret (Ardea alba), Great Blue Heron (Ardea herodias), Snowy Egret (Egretta thula), Wood Stork (Mycteria americana), Roseate Spoonbill (Platalea ajaja), and Anhinga (Anhinga anhinga). These species are widely geographically distributed and therefore likely to occur in airborne imagery around the world. They also represent a diverse range of large wading bird appearances, making the dataset well suited to pre-training models for use in other ecosystems. This dataset was initially produced to support the development of computer vision models for monitoring one of the largest ongoing wetland restoration efforts in the world using wading birds as indicators. It has also been used to pre-train a key general bird detection model. The dataset will be useful for continued improvement of models for population and community monitoring of the Everglades in the United States (a crucial ecosystem undergoing large-scale restoration), as training data for airborne models involving widely distributed bird species, and will serve as a key data source for pre-training other bird detection and classification models (due to its size and mixed-species nature), thus supporting the development of individual-level airborne computer vision approaches more broadly. The data are available for reuse under CC0 1.0 Universal Licensing.