Skip to main content
. 2021 Aug 10;11:16244. doi: 10.1038/s41598-021-95545-y

Table 6.

The architecture of the proposed deep learning network.

Layer Features (train) Features (inference) Kernel size Stride
Input 512 × 512 × 3 512 × 512 × 3
Conv1_1 512 × 512 × 3 512 × 512 × 3 3 × 3 1 × 1
Conv1_2 710 × 710 × 64 710 × 710 × 64 3 × 3 1× 1
Pool1 710 × 710 × 64 710 × 710 × 64 2 × 2 2 × 2
Conv2_1 355 × 355 × 64 355 × 355 × 64 3 × 3 1 × 1
Conv2_2 355 × 355 × 128 355 × 355 × 128 3 × 3 1 × 1
Pool2 355 × 355 × 128 355 × 355 × 128 2 × 2 2 × 2
Conv3_1 178 × 178 × 128 178 × 178 × 128 3 × 3 1 × 1
Conv3_2 178 × 178 × 256 178 × 178 × 256 3 × 3 1 × 1
Conv3_3 178 × 178 × 256 178 × 178 × 256 3 × 3 1 × 1
Pool3 178 × 178 × 256 178 × 178 × 256 2 × 2 2 × 2
Conv4_1 89 × 89 × 256 89 × 89 × 256 3 × 3 1 × 1
Conv4_2 89 × 89 × 512 89 × 89 × 512 3 × 3 1 × 1
Conv4_3 89 × 89 × 512 89 × 89 × 512 3 × 3 1 × 1
Pool4 89 × 89 × 512 89 × 89 × 512 2 × 2 2 × 2
Conv5_1 45 × 45 × 512 45 × 45 × 512 3 × 3 1 × 1
Conv5_2 45 × 45 × 512 45 × 45 × 512 3 × 3 1 × 1
Conv5_3 45 × 45 × 512 45 × 45 × 512 3 × 3 1 × 1
Pool5 45 × 45 × 512 45 × 45 × 512 2 × 2 2 × 2
Drop6 23 × 23 × 512 16 × 16 × 512
Drop7 17 × 17 × 4096 10 × 10 × 4096
Upsampled 576 × 576 × 3 576 × 576 × 3
Output 512 × 512 × 3 512 × 512 × 3