深度学总结：CNN Decoder, Upsampling的处理

CNN Decoder, 需要做Upsampling：

金字塔结构收缩的dimensions要膨胀回来，处理方法就是Upsampling，直接复制（容易产生棋盘条纹），或者做内插interpolation，然后再做convolution：

# define the NN architecture
class ConvAutoencoder(nn.Module):
    def __init__(self):
        super(ConvAutoencoder, self).__init__()
        ## encoder layers ##
        # conv layer (depth from 1 --> 16), 3x3 kernels
        self.conv1 = nn.Conv2d(1, 16, 3, padding=1)  
        # conv layer (depth from 16 --> 8), 3x3 kernels
        self.conv2 = nn.Conv2d(16, 4, 3, padding=1)
        # pooling layer to reduce x-y dims by two; kernel and stride of 2
        self.pool = nn.MaxPool2d(2, 2)
        
        ## decoder layers ##
        self.conv4 = nn.Conv2d(4, 16, 3, padding=1)
        self.conv5 = nn.Conv2d(16, 1, 3, padding=1)
        

    def forward(self, x):
        # add layer, with relu activation function
        # and maxpooling after
        x = F.relu(self.conv1(x))
        x = self.pool(x)
        # add hidden layer, with relu activation function
        x = F.relu(self.conv2(x))
        x = self.pool(x)  # compressed representation
        
        ## decoder 
        # upsample, followed by a conv layer, with relu activation function  
        # this function is called `interpolate` in some PyTorch versions
        x = F.upsample(x, scale_factor=2, mode='nearest')
        x = F.relu(self.conv4(x))
        # upsample again, output should have a sigmoid applied
        x = F.upsample(x, scale_factor=2, mode='nearest')
        x = F.sigmoid(self.conv5(x))
        
        return x

pytorch包含了nn.ConvTranspose2d直接处理的办法：

# define the NN architecture
class ConvAutoencoder(nn.Module):
    def __init__(self):
        super(ConvAutoencoder, self).__init__()
        ## encoder layers ##
        # conv layer (depth from 1 --> 16), 3x3 kernels
        self.conv1 = nn.Conv2d(1, 16, 3, padding=1)  
        # conv layer (depth from 16 --> 4), 3x3 kernels
        self.conv2 = nn.Conv2d(16, 4, 3, padding=1)
        # pooling layer to reduce x-y dims by two; kernel and stride of 2
        self.pool = nn.MaxPool2d(2, 2)
        
        ## decoder layers ##
        ## a kernel of 2 and a stride of 2 will increase the spatial dims by 2
        self.t_conv1 = nn.ConvTranspose2d(4, 16, 2, stride=2)
        self.t_conv2 = nn.ConvTranspose2d(16, 1, 2, stride=2)


    def forward(self, x):
        ## encode ##
        # add hidden layers with relu activation function
        # and maxpooling after
        x = F.relu(self.conv1(x))
        x = self.pool(x)
        # add second hidden layer
        x = F.relu(self.conv2(x))
        x = self.pool(x)  # compressed representation
        
        ## decode ##
        # add transpose conv layers, with relu activation function
        x = F.relu(self.t_conv1(x))
        # output layer (with sigmoid for scaling from 0 to 1)
        x = F.sigmoid(self.t_conv2(x))
                
        return x