Cutting the visual world into bigger slices for improved video concept detection