models.py

import torch
import torch.nn as nn
from torch.autograd import Variable
#weight initialisation with mean=0 and stddev=0.02

device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')

def weights_init(m):
    classname = m.__class__.__name__
    if classname.find('Conv') != -1:
        nn.init.normal_(m.weight.data, 0.0, 0.02)
    elif classname.find('BatchNorm') != -1:
        nn.init.normal_(m.weight.data, 1.0, 0.02)
        nn.init.constant_(m.bias.data, 0)

class Encoder(nn.Module):
  def __init__(self,in_channels,latent_dim):              #in_channels=1 for MNIST and in_channels=3 for CIFAR10
    super(Encoder,self).__init__()
    self.conv1=nn.Conv2d(in_channels,64,5,padding=2,stride=2)   
    self.bn1=nn.BatchNorm2d(64,momentum=0.9)
    self.conv2=nn.Conv2d(64,128,5,padding=2,stride=2)
    self.bn2=nn.BatchNorm2d(128,momentum=0.9)
    self.conv3=nn.Conv2d(128,256,5,padding=2,stride=2)
    self.bn3=nn.BatchNorm2d(256,momentum=0.9)
    self.relu=nn.LeakyReLU(0.2)
    self.fc1=nn.Linear(256*8*8,2048)
    self.bn4=nn.BatchNorm1d(2048,momentum=0.9)
    self.fc_mean=nn.Linear(2048,latent_dim)
    self.fc_logvar=nn.Linear(2048,latent_dim)   #latent dim=128
  
  def forward(self,x):
    batch_size=x.size()[0]
    out=self.relu(self.bn1(self.conv1(x)))
    out=self.relu(self.bn2(self.conv2(out)))
    out=self.relu(self.bn3(self.conv3(out)))
    out=out.view(batch_size,-1)
    out=self.relu(self.bn4(self.fc1(out)))
    mean=self.fc_mean(out)
    logvar=self.fc_logvar(out)
    
    return mean,logvar

class Decoder(nn.Module):
  def __init__(self,latent_dim,out_channels):
    super(Decoder,self).__init__()
    self.fc1=nn.Linear(latent_dim,8*8*256)
    self.bn1=nn.BatchNorm1d(8*8*256,momentum=0.9)
    self.relu=nn.LeakyReLU(0.2)
    self.deconv1=nn.ConvTranspose2d(256,256,6, stride=2, padding=2)
    self.bn2=nn.BatchNorm2d(256,momentum=0.9)
    self.deconv2=nn.ConvTranspose2d(256,128,6, stride=2, padding=2)
    self.bn3=nn.BatchNorm2d(128,momentum=0.9)
    self.deconv3=nn.ConvTranspose2d(128,32,6, stride=2, padding=2)
    self.bn4=nn.BatchNorm2d(32,momentum=0.9)
    self.deconv4=nn.ConvTranspose2d(32,out_channels,5, stride=1, padding=2)
    self.tanh=nn.Tanh()

  def forward(self,x):
    batch_size=x.size()[0]
    x=self.relu(self.bn1(self.fc1(x)))
    x=x.view(-1,256,8,8)
    x=self.relu(self.bn2(self.deconv1(x)))
    x=self.relu(self.bn3(self.deconv2(x)))
    x=self.relu(self.bn4(self.deconv3(x)))
    x=self.tanh(self.deconv4(x))
    return x

class Discriminator(nn.Module):
  def __init__(self,in_channels):
    super(Discriminator,self).__init__()
    self.conv1=nn.Conv2d(in_channels,32,5,padding=2,stride=1)
    self.relu=nn.LeakyReLU(0.2)
    self.conv2=nn.Conv2d(32,128,5,padding=2,stride=2)
    self.bn1=nn.BatchNorm2d(128,momentum=0.9)
    self.conv3=nn.Conv2d(128,256,5,padding=2,stride=2)
    self.bn2=nn.BatchNorm2d(256,momentum=0.9)
    self.conv4=nn.Conv2d(256,256,5,padding=2,stride=2)
    self.bn3=nn.BatchNorm2d(256,momentum=0.9)
    self.fc1=nn.Linear(8*8*256,512)
    self.bn4=nn.BatchNorm1d(512,momentum=0.9)
    self.fc2=nn.Linear(512,1)
    self.sigmoid=nn.Sigmoid()

  def forward(self,x):
    batch_size=x.size()[0]
    x=self.relu(self.conv1(x))
    x=self.relu(self.bn1(self.conv2(x)))
    x=self.relu(self.bn2(self.conv3(x)))
    x=self.relu(self.bn3(self.conv4(x)))
    x=x.view(-1,256*8*8)
    x1=x;
    x=self.relu(self.bn4(self.fc1(x)))
    x=self.sigmoid(self.fc2(x))

    return x,x1

class VAE_GAN(nn.Module):
  def __init__(self,in_channels,out_channels,latent_dim):
    super(VAE_GAN,self).__init__()
    self.encoder=Encoder(in_channels,latent_dim)
    self.decoder=Decoder(latent_dim,out_channels)
    self.discriminator=Discriminator(in_channels)
    self.encoder.apply(weights_init)
    self.decoder.apply(weights_init)
    self.discriminator.apply(weights_init)


  def forward(self,x):
    bs=x.size()[0]
    z_mean,z_logvar=self.encoder(x)
    std = z_logvar.mul(0.5).exp_()
    l_dim=z_mean.size()[1]    
    #sampling epsilon from normal distribution
    epsilon=Variable(torch.randn(bs,l_dim)).to(device)
    z=z_mean+std*epsilon
    x_tilda=self.decoder(z)
      
    return z_mean,z_logvar,x_tilda