Semantic Self-adaptation: Enhancing Generalization with a Single Sample

The lack of out-of-domain generalization is a critical weakness of deep networks for semantic segmentation. Previous studies relied on the assumption of a static model, i. e.,...