Aquaculture Dataset: Black Tiger Prawn Segmentation

Understanding the growth and distribution of the prawns is critical for optimising the feed and harvest strategies. However, the most commonly adopted sampling practice, the cast net approach, is unable to sample the prawns at a high frequency as it is expensive and laborious. An alternative approach is to use computer vision techniques to sample prawns from feed trays that farm workers inspect multiple times each day. \n\nBeing able to accurately detect and segment prawns is the very first step towards frequent prawn growth monitoring using computer vision techniques. However, there is no publicly available prawn segmentation data. This dataset is the very first dataset that provides polygon annotations (COCO format) for Black Tiger Prawns (Penaeus monodon).\nLineage: The raw data was collected from four research ponds at CSIRO Bribie Island Research Station using an Intel RealSense D435i camera between October and December 2021. The camera was mounted on a helmet that farm technicians wore. The raw dataset was recorded in Robot Operating System (ROS) Bag format, which includes a colour stream (BGR8, 1280x720, 15FPS). The datasets contain RGB images of a variety of lighting/weather conditions. Each RGB image contains prawns of various sizes.\n\nThis real-world data collection contains 735 randomly selected RGB images of size 1280x720. In total, 4454 prawns were annotated as polygons in the MS COCO format.