DAVIS: Densely Annotated VIdeo Segmentation