Learning intra-domain style-invariant representation for unsupervised domain adaptation of semantic segmentation

作者：

Highlights：

• A novel conception of learning intra-domain style-invariant representation for unsupervised domain adaptation of semantic segmentation.

• A self-ensembling method for learning the intra-domain style-invariant representation.

• A semantic-aware multimodal image-to-image translation model for obtaining images with diversified intra-domain styles.

• Extensive experiments and analyses for validating the effectiveness and superiority over state-of-the-art methods.

摘要

•A novel conception of learning intra-domain style-invariant representation for unsupervised domain adaptation of semantic segmentation.•A self-ensembling method for learning the intra-domain style-invariant representation.•A semantic-aware multimodal image-to-image translation model for obtaining images with diversified intra-domain styles.•Extensive experiments and analyses for validating the effectiveness and superiority over state-of-the-art methods.

论文关键词：Style-invariant representation,Self-ensembling,Domain adaptation

论文评审过程：Received 19 May 2021, Revised 21 June 2022, Accepted 17 July 2022, Available online 20 July 2022, Version of Record 25 July 2022.

论文官网地址：https://doi.org/10.1016/j.patcog.2022.108911