A Spatiotemporal Mask Autoencoder for One-shot Video Object Segmentation