Super-resolution semantic segmentation with relation calibrating network

作者：

Highlights：

• A Relation Calibrating Network (RCNet) for the task of super-resolution semantic segmentation is proposed to alleviate the computational burden of general semantic segmentation network.

• A relation upsampling module is proposed to model the correlations between pixels and their neighboring pixels for relation propagation.

• A feature calibrating module is introduced to calibrate high-resolution features which upsampled from low-resolution features.

• RCNet achieves impressive results on Cityscapes and PASCAL Context datasets.

摘要

•A Relation Calibrating Network (RCNet) for the task of super-resolution semantic segmentation is proposed to alleviate the computational burden of general semantic segmentation network.•A relation upsampling module is proposed to model the correlations between pixels and their neighboring pixels for relation propagation.•A feature calibrating module is introduced to calibrate high-resolution features which upsampled from low-resolution features.•RCNet achieves impressive results on Cityscapes and PASCAL Context datasets.

论文关键词：Image semantic segmentation,Super-resolution semantic segmentation,Relation calibrating

论文评审过程：Received 22 February 2021, Revised 13 December 2021, Accepted 18 December 2021, Available online 21 December 2021, Version of Record 3 January 2022.

论文官网地址：https://doi.org/10.1016/j.patcog.2021.108501