Domain adaptation with contrastive and adversarial oriented transferable semantic augmentation

Wenxu Wang, Guoxu Zhang, Ping Zhong

Abstract: Highlights•We propose a novel UDA method by implicit semantic augmentation.•It learns the discrimination features by contrastive semantic augmentation strategy.•It learns the domain-invariant features by adversarial semantic augmentation manner.•It does not need to generate actual images and saves overhead.•It can be easily embedded into other UDA methods to benefit each other.