Long-tail Video Object Segmentation