Automated fish measurement and speciation has the potential to revolutionize the fishery monitoring process, driving down costs and reducing the burden on human observers and video reviewers. Many efforts are underway to apply machine learning algorithms to different problems related to fisheries science. These efforts could benefit greatly from the large open data science community, exemplified by sites like Kaggle and DrivenData. These sites provide a platform to host an open data set and offer a prize for contestants to solve challenging machine learning problems. Using a grant from the National Fish & Wildlife Foundation, our goal was to create a high-quality data set to push forward the state of the art in automatic processing of Electronic Monitoring footage. We obtained footage from the New England groundfishing fleet collected over several seasons with help from the Greater Maine Research Institute. We then developed a software platform to analyze, clean, and annotate videos for use in an open data science competition. Using these videos, we hosted a data science competition, resulting in open source algorithms available to researchers and industry alike.
