Code, imagery, and annotations for training a deep learning model to detect wildlife in aerial imagery

There are 3 child zip files included in this data release. 01_Codebase.zip contains a codebase for using deep learning to filter images based on the probability of any bird occurrence. It includes instructions and files necessary for training, validating, and testing a machine learning detection algorithm. 02_Imagery.zip contains imagery that were collected using a Partenavia P68 fixed-wing airplane using a PhaseOne iXU-R 180 forward motion compensating 80-megapixel digital frame camera with a 70 mm Rodenstock lens. The imagery were cropped into smaller patches of 720x720 pixels for training and 1440x1440 pixels for validation and test datasets. These data were collected for developing machine learning algorithms for the detection and classification of avian targets in aerial imagery. These data can be paired with annotation values to train and evaluate object detection and classification models. 03_Annotations.zip contains a collection of bounding boxes around avian targets in aerial imagery formatted as COCO JSON file. The data are nested under evaluation and test folders and contain both ground truth targets and predicted targets.These data were collected for two main functions. The ground truth avian targets were manually annotated and can be used to train avian detection algorithms using machine learning methods. The predicted targets can be used to evaluate model performance while referencing the larger work associated with these data.