To recognize the target classes with only a few samples, few-shot learning (FSL) uses prior knowledge learned from the source classes and is usually expressed as a special domain adaptation problem. However, Existing few-shot learning methods make the implicit assumption that the few target class samples are from the same domain or different domain as the source class samples, which greatly limits their application in the wild. This paper introduces a few-shot learning method in multiple scenarios which requires no prior knowledge on the label set. For a given target domain labels set, it may overlap with the set of source domain labels to varying degrees, thereby bringing up an additional class gap based on the domain gap. In order to solve this problem in a unified framework, we propose a novel domain adaptation network which is designed to address a specific challenge: How to achieve domain adaptation whilst maintaining source/target per-class discriminativeness when the target domain label set is unseen. Our solution is to design corresponding soft label transfer network and minimax entropy network for different target domain label set, then we quantify the transferability of the source domain label set during the training process to discover the relationship between the source domain label set and the target domain label set. Further, we broaden the model's understanding of the data by learning from self-supervised signals how to solve a jigsaw puzzle on the same sample. Extensive experiments show that our model outperforms the state-of-the-art models.