MovieCuts: A New Dataset and Benchmark for Cut Type Recognition

Understanding movies and their structural patterns is a crucial task in decoding the craft of video editing. While previous works have developed tools for general analysis, such as detecting characters or recognizing cinematography properties at shot level, less effort has been devoted to understanding most basic edit, Cut. This paper introduces Cut type recognition task, which requires modeling multi-modal information. To ignite research this new we construct large-scale dataset called MovieCuts, contains 173, 967 clips labeled with ten cut types defined by professionals movie industry. We benchmark set audio-visual approaches, including some dealing problem’s nature. Our best model achieves $$47.7\%$$ mAP, suggests that challenging attaining highly accurate an open problem. Advances automatic Cut-type can unleash experiences editing industry, analysis education, re-editing, virtual cinematography, machine-assisted trailer generation, editing, among others. data code are publicly available: https://github.com/PardoAlejo/MovieCuts .