SODA: A Dataset for Small Object Detection in UAV Captured Imagery

Modern Unmanned Aerial Vehicles (UAVs) are equipped with high-resolution cameras that can capture imagery and video of whatever lies in their field of view. However, depending on the altitude of the UAV, objects within the imagery appear small, sometimes covering just a few pixels. This makes the detection of such objects very challenging. The SODA (Small Objects at Different Altitudes) dataset was created to assist in research dealing with small object detection from aerial imagery. The dataset contains 377 aerial images captured using UAVs at altitudes ranging between 5m and 30 m. The dataset is then used to evaluate the performance of YOLOv8. The SODA dataset is publicly available on GitHub¹.