This page introduces the Object detection dataset we created in our work. It contains 1000 images with all vessel-like objects annotated by humans. The dataset is provided in COCO format.
Our github repository contains further information about the object detection dataset under the path "datasets/2D_vessel_detection_coco". The github contains a heavily compressed version of the dataset, applying strong jpeg compression to the images. Object detectors can handle this compressed dataset, but perform worse then on the original dataset. Next to the compressed zip, there is a truncated version of the original, showing a few images that where used in our work with their native quality.
The full, native version of the dataset is available in the UHH cloud.