Skip to main content
. 2022 Nov 30;24(12):1755. doi: 10.3390/e24121755

Figure 2.

Figure 2

The structure of the proposed network. backbone: Res2Net50 that is pre-trained on ImageNet dataset. HDM: high-level feature decoder module. It fully aggregates multi-scale high-level features to generate rough global features. LDM: low-level feature decoder module. The generated edge features are expected to refine and constrain the coarse global features. CM: complementary module. It complements the edge and global features to obtain a segmentation prediction map. LE: low-level edge block. MF: multi-receptive field block. MD: multi-scale decoder block. LF: location fusion. EF: edge fusion.