The EyeSea underwater video dataset was assembled for developing algorithms for detecting fish in real world underwater video data. The data were recorded as part of environmental monitoring efforts at three different water power sites. The Ocean Renewable Power Company (ORPC) data were recorded in the Kvijak River in Alaska, where ORPC deployed their prototype RivGen in-stream turbine in 2015. The Voith Hydro data were recorded by Aquatera in Fall of Warness, Scotland during the deployment of the Voith Hydro 1 MW HyTide tidal energy converter. The Wells Dam data were recorded at Douglas County Public Utility District’s hydropower dam on the Columbia River in Washington. Each of these three sites represents a different real-world underwater environment typical of water power generation. The data were annotated by staff at PNNL with bounding boxes around the location of fish in each video frame. The data is structured for use as training and test data for machine learning in the Tensorflow Keras environment, as described in the referenced paper.
Training and validating (train_x, train_y)
Testing (voith_test_x, voith_test_y, wells_test_x, wells_test_y, orpc_test_x, orpc_test_y)
Original videos (under videos, Aquatera_VoithHydro, DCPUD_WellsDam, ORPC_Igiugig)