Skip to main content
. 2022 Sep 1;16:971829. doi: 10.3389/fnins.2022.971829

Table 1.

Architectures of the proposed generator and discriminator.

Layers Output shape Norm./Act.
Generator
Input 2 ×128 ×128
Padding 2 ×128 ×128 ELR/LReLU/PN
Conv2d 32 ×64 ×64 ELR/LReLU/PN
Conv2d 64 ×32 ×32 ELR/LReLU/PN
ResBlock*2 64 ×32 ×32 ELR/LReLU/PN
Upsample 64 ×64 ×64
Conv2d 32 ×64 ×64 ELR/LReLU/PN
Upsample 32 ×128 ×128
Conv2d 16 ×128 ×128 ELR/LReLU/PN
Padding 16 ×134 ×134
Conv2d 2 ×128 ×128 Tanh
Discriminator
Input 4 ×128 ×128
Conv2d 16 ×64 ×64 ELR/LReLU
Conv2d 32 ×32 ×32 ELR/LReLU
Conv2d 64 ×16 ×16 ELR/LReLU
Conv2d 128 ×15 ×15 ELR/LReLU
Conv2d 1 ×14 ×14