Text this: Learning discriminative foreground-and-background features for few-shot segmentation.