A deep learning dataset for savanna tree species in Northern Australia

We present a baseline deep learning dataset of 2547 polygons for 36 tree species in Northern Australia. Polygons were drawn on imagery that was collected using Remotely Piloted Aircraft System (RPAS). The dataset consists of: 7 orthomosaics 7 shape files with polygon annotations 1 training dataset in COCO format 1 validation dataset in COCO format Training and validation datasets were derived from the orthomosaics by tiling each image at 1024x1024 pixel size with 512 pixel step size (overlap). To perform deep learning model training with this dataset go to https://github.com/ajansenn/SavannaTreeAI for more information.