语义分割--Label Refinement Network for Coarse-to-Fine Semantic Segmentation

Label Refinement Network for Coarse-to-Fine Semantic Segmentation

https://www.arxiv.org/abs/1703.00551v1

还是级联语义分割的思路，本文提出的算法是 label refinement network (LRN)

首先来看看本文提出的算法流程图：
语义分割--Label Refinement Network for Coarse-to-Fine Semantic Segmentation
注意这里的训练加入了多尺度 ground truth

refinement module (RE)
语义分割--Label Refinement Network for Coarse-to-Fine Semantic Segmentation
这个模块的两个输入：一个 coarse label map，一个是 skip feature map。
注意这里输入输出的尺寸变化

语义分割--Label Refinement Network for Coarse-to-Fine Semantic Segmentation

PASCAL VOC 2012
语义分割--Label Refinement Network for Coarse-to-Fine Semantic Segmentation

中间各个步骤结果：
语义分割--Label Refinement Network for Coarse-to-Fine Semantic Segmentation

CamVid dataset 结果：
语义分割--Label Refinement Network for Coarse-to-Fine Semantic Segmentation

语义分割--Label Refinement Network for Coarse-to-Fine Semantic Segmentation

SUN RGB-D dataset
语义分割--Label Refinement Network for Coarse-to-Fine Semantic Segmentation

语义分割--Label Refinement Network for Coarse-to-Fine Semantic Segmentation