Generalizable model-agnostic semantic segmentation via target-specific normalization

作者：

Highlights：

• The method is designed for the generalizable semantic segmentation task.

• The method trains model-agnostic model with meta learning.

• The method adapts to the testing domain with target-specific normalization.

• Image bank with the style-based selection policy obtains more accurate statistics.

摘要

•The method is designed for the generalizable semantic segmentation task.•The method trains model-agnostic model with meta learning.•The method adapts to the testing domain with target-specific normalization.•Image bank with the style-based selection policy obtains more accurate statistics.

论文关键词：Domain generalization,Semantic segmentation,Model-agnostic learning,Target-specific normalization

论文评审过程：Received 1 March 2021, Revised 13 August 2021, Accepted 30 August 2021, Available online 8 September 2021, Version of Record 25 September 2021.

论文官网地址：https://doi.org/10.1016/j.patcog.2021.108292