Optimized HRNet for image semantic segmentation

作者：

Highlights：

• An optimized High-Resolution Net (HRNet) for image semantic segmentation.

• A novel MDC module is proposed to improve multi-scale segmentation.

• A novel MDFA module is proposed to optimize multi-level feature fusion.

• A data-independent upsampling scheme is proposed to recover pixel-level prediction.

摘要

•An optimized High-Resolution Net (HRNet) for image semantic segmentation.•A novel MDC module is proposed to improve multi-scale segmentation.•A novel MDFA module is proposed to optimize multi-level feature fusion.•A data-independent upsampling scheme is proposed to recover pixel-level prediction.

论文关键词：Image semantic segmentation,High-resolution net,Mixed dilated convolution,Multi-level data-dependent feature aggregation

论文评审过程：Received 18 July 2019, Revised 2 December 2020, Accepted 19 December 2020, Available online 29 December 2020, Version of Record 11 March 2021.

论文官网地址：https://doi.org/10.1016/j.eswa.2020.114532