model/TransformerNet.py

# import package : model
import torch
from torch import nn
from torch.nn import functional as F

class ConvLayer(nn.Module):
    def __init__(self, in_channels, out_channels, kernel_size, stride):
        super(ConvLayer, self).__init__()
        reflection_padding = kernel_size // 2
        self.reflection_pad = nn.ReflectionPad2d(reflection_padding)
        self.conv2d = nn.Conv2d(in_channels, out_channels, kernel_size, stride)

    def forward(self, x):
        out = self.reflection_pad(x)
        out = self.conv2d(out)
        return out
    
class ResidualBlock(nn.Module):
    """ResidualBlock
    introduced in: https://arxiv.org/abs/1512.03385
    recommended architecture: http://torch.ch/blog/2016/02/04/resnets.html
    """

    def __init__(self, channels):
        super(ResidualBlock, self).__init__()
        self.conv1 = ConvLayer(channels, channels, kernel_size=3, stride=1)
        self.in1 = nn.InstanceNorm2d(channels, affine=True)
        self.conv2 = ConvLayer(channels, channels, kernel_size=3, stride=1)
        self.in2 = nn.InstanceNorm2d(channels, affine=True)
        self.relu = nn.ReLU()

    def forward(self, x):
        residual = x
        out = self.relu(self.in1(self.conv1(x)))
        out = self.in2(self.conv2(out))
        out = out + residual
        return out
    
class TransformerNet(nn.Module):
    def __init__(self):
        super(TransformerNet, self).__init__()

        # Encoder convolution layers
        self.encoder = nn.Sequential()
        
        self.encoder.add_module('conv1', ConvLayer(3, 32, kernel_size=9, stride=1))
        self.encoder.add_module('in1', nn.InstanceNorm2d(32, affine=True))
        self.encoder.add_module('relu1', nn.ReLU())

        # Residual layers
        self.residual = nn.Sequential()
        
        for i in range(5):
            self.residual.add_module('resblock_%d' %(i+1), ResidualBlock(32))
        
        # Decoder Layers
        self.decoder = nn.Sequential()

        self.decoder.add_module('deconv3', ConvLayer(32, 3, kernel_size=9, stride=1))


    def forward(self, x):
        encoder_output = self.encoder(x)
        residual_output = self.residual(encoder_output)
        decoder_output = self.decoder(residual_output)
        
        return decoder_output