论文地址

第一阶段

conv2d(3×3)

first conv：572×572×1 → 570×570×64

second conv：570×570×64 → 568×568×64

代码

# first 3×3 convolutional layer
self.first = nn.Conv2d(in_channels, out_channels, kernel_size=3, padding=1)
self.act1 = nn.ReLU()
# Second 3×3 convolutional layer
self.second = nn.Conv2d(out_channels, out_channels, kernel_size=3, padding=1)
self.act2 = nn.ReLU()

第二阶段

maxpool2d

568×568×64 → 284×284×128

self.pool = nn.MaxPool2d(2)

conv2d

...

第五阶段

maxpool2d

64×64×512 → 32×32×1024

conv2d

→ 30×30×1024 → 28×28×1024

未完待补充