Scene-driven Retrieval in Edited Videos using Aesthetic and Semantic Deep Features