Domain generalization and adaptation based on second-order style information

作者：

Highlights：

• We proposed a two-level style normalization and restitution (Tl-SNR) for both domain generalization and unsupervised domain adaptation.

• We extend the theory of group whitening to the domain of domain generalization and unsupervised domain adaptation.

• We defined dynamic affine parameters to improve group whitening

• The methods achieve state-of-the-art performance on the general DG and UDA benchmarks.

摘要

•We proposed a two-level style normalization and restitution (Tl-SNR) for both domain generalization and unsupervised domain adaptation.•We extend the theory of group whitening to the domain of domain generalization and unsupervised domain adaptation.•We defined dynamic affine parameters to improve group whitening•The methods achieve state-of-the-art performance on the general DG and UDA benchmarks.

论文关键词：Domain generalization,Unsupervised domain adaptation,Two-level style normalization and restitution,Second-order statistics,Dynamic affine parameter

论文评审过程：Received 17 May 2021, Revised 21 December 2021, Accepted 16 February 2022, Available online 3 March 2022, Version of Record 9 March 2022.

论文官网地址：https://doi.org/10.1016/j.patcog.2022.108595