Tilapia-RAS Dataset: Underwater Videos and Polygon-Annotated Frames with Physicochemical Metadata

This dataset provides a real-world collection of underwater videos featuring Nile tilapia (Oreochromis niloticus) within a commercial recirculating aquaculture system (RAS). The dataset comprises 31 curated 30-second video clips , captured under realistic production challenges, including variable turbidity, suspended solids, and diverse illumination configurations (natural, frontal, and back-mounted lighting).

Key resources include:

• A structured CSV metadata file (meta_tilapia_set.csv) providing synchronized physicochemical water quality parameters (temperature, pH, dissolved oxygen, and turbidity) for each video clip.
• A subset of 3,520 extracted frames from four fully annotated clips. Annotations are provided as polygon instance masks in LabelMe (JSON) format.

This dataset is intended to support the development and benchmarking of robust computer vision models for non-invasive aquaculture monitoring, including object detection, segmentation, and behavior recognition tasks