VOST: Video Object Segmentation under Transformations