UAV Thermal Dataset for Wildlife Detection: Focusing on Koala

Dataset DescriptionThis dataset contains thermal drone imagery of koalas, collected in dense eucalyptus forest environments in Australia. Images were captured at a resolution of 640×512 pixels using drone-mounted thermal sensors, designed to address the challenges of ecological monitoring under canopy cover and low-contrast conditions.The dataset includes 4,277 training images and 1,252 validation images, annotated across two categories:Koala – confirmed thermal detections of koalas.Non-Koala – other heat signatures (e.g., vegetation, branches, or other non-target animals).Annotations are provided in both YOLO .txt format and COCO-style .json format, ensuring compatibility with widely used object detection frameworks such as YOLOv5/YOLOv8, MMDetection, and Detectron2.Folder StructureKoala/├── images/ # Thermal drone image frames├── annotations_yolo/ # YOLO-format annotations├── annotations_coco/ # COCO-style JSON annotations├── README_Koala.md # Dataset-specific documentationAnnotation DetailsYOLO Format: [class_id x_center y_center width height] with all coordinates normalized to image size.COCO Format: Standard images, annotations, and categories fields compatible with pycocotools and other libraries.Potential ApplicationsWildlife monitoring and conservation – improving automated detection of koalas for ecological studies.Small object detection in thermal UAV imagery – benchmarking object detection under canopy occlusion.Model generalization studies – evaluating transferability and domain adaptation for thermal datasets.AI for conservation – supporting tools to aid field researchers in real-time koala population surveys.