The pretraining-finetuning paradigm has gained widespread adoption in vision tasks and other fields. However, the finetuning phase still requires high-quality annotated samples. To overcome this ...