Abstract:
Domain adaptation aims to address the challenge of transferring the knowledge obtained from the source domain with rich label information to the target domain with less or even no label information. Recent methods start to tackle this problem by incorporating the hard-pseudo labels for the target samples to better reduce the cross-domain distribution shifts. However, these approaches are vulnerable to the error accumulation and hence unable to preserve cross-domain category consistency. Because the accuracy of pseudo labels cannot be guaranteed explicitly. To address this issue, we propose a Soft Labels transfer with Discriminative Representations learning (SLDR) framework to jointly optimize the class-wise adaptation with soft target labels and learn the discriminative domain-invariant features in a unified model. Specifically, to benefit each other in an effective manner, we simultaneously explore soft target labels by label propagation for better condition adaptation and preserve the important properties of inter-class separability and intra-class compactness for reducing more domain shifts. Extensive experiments are conducted on several unsupervised domain adaptation datasets, and the results show that SLDR outperforms the state-of-the-art methods.
