This labeled dataset contains 107,451 acoustic camera video frames capturing marine life interactions around an underwater tidal turbine. Each frame is annotated with bounding boxes identifying marine life objects as labeled by a fish biologist. Created to support research into automated target detection around underwater turbines, this dataset aims to advance capabilities that enable the safe deployment and operation of marine energy devices. No collisions were observed with the turbine while labeling and analyzing this dataset, and a publication detailing this new work will be added once available. The video data was originally collected in 2010 around Ocean Renewable Power Company's (ORPC) tidal turbine deployment in Cobscook Bay, Maine, USA, with results published in Viehman and Zydlewski (2015) Estuaries and Coasts 38: 241-252 (linked below). Code, software tools, and a baseline automated detection approach developed for this effort are available in the PNNL-TUNAMELT GitHub repository, which also provides guidance for getting started with this dataset. For further information, please refer to the GitHub repository, the associated publication, or contact the authors.