I’ll be happy to give you a concrete code example or architecture suggestion for extracting deep features from that video.