---

Avisek Gupta

Indian Statistical Institute, Kolkata

---

---

Tutorial 4: Deep Convolution Generative Adversarial Networks

---

import torch
import torch.nn as nn
import torch.optim as optim
import torch.nn.functional as F
import torchvision
import torchvision.transforms as transforms
import numpy as np
import matplotlib.pyplot as plt


class Generator(nn.Module):
    def __init__(self, noise_dim):
        super(Generator, self).__init__()
        self.convT1 = nn.ConvTranspose2d(noise_dim, 128, 4, 1, 0)
        self.convT2 = nn.ConvTranspose2d(128, 64, 3, 2, 1)
        self.convT3 = nn.ConvTranspose2d(64, 32, 4, 2, 1)
        self.convT4 = nn.ConvTranspose2d(32, 1, 4, 2, 1)

    def forward(self, input):
        x = F.relu(self.convT1(input))
        x = F.relu(self.convT2(x))
        x = F.relu(self.convT3(x))
        x = torch.tanh(self.convT4(x))
        return x


class Discriminator(nn.Module):
    def __init__(self):
        super(Discriminator, self).__init__()
        self.conv1 = nn.Conv2d(1, 32, 4, 2, 1)
        self.conv2 = nn.Conv2d(32, 64, 4, 2, 1)
        self.conv3 = nn.Conv2d(64, 128, 3, 2, 1)
        self.conv4 = nn.Conv2d(128, 1, 4, 1, 0)

    def forward(self, input):
        x = F.leaky_relu(self.conv1(input), negative_slope=0.2)
        x = F.leaky_relu(self.conv2(x), negative_slope=0.2)
        x = F.leaky_relu(self.conv3(x), negative_slope=0.2)
        x = self.conv4(x)
        return x


if __name__ == '__main__':
    
    workers = 4 # Number of workers for dataloader
    batch_size = 128
    noise_dim = 100 # Size of z latent vector
    num_epochs = 100

    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    print('Using device:', device)

    dataset = torchvision.datasets.MNIST(
        root='./data', train=True, download=True, 
        transform=transforms.Compose([
            transforms.ToTensor(), transforms.Lambda(lambda x: ((x * 2) - 1))
        ])
    )
    
    dataloader = torch.utils.data.DataLoader(
        dataset, batch_size=batch_size, shuffle=True, num_workers=workers
    )
    
    netG = Generator(noise_dim).to(device)
    
    netD = Discriminator().to(device)
    

    # convention for real and fake labels
    real_label = 1.0
    fake_label = 0.0

    criterion = nn.BCEWithLogitsLoss()
    optimizerD = optim.Adam(netD.parameters(), lr=0.002, betas=(0.5, 0.999))
    optimizerG = optim.Adam(netG.parameters(), lr=0.001, betas=(0.5, 0.999))

    for epoch in range(num_epochs):
        ep_G_loss = 0
        ep_D_loss = 0
        for i, data in enumerate(dataloader, 0):
            realX = data[0].to(device)
            b_size = realX.size(0)

            # Update D: maximize log(D(x)) + log(1 - D(G(z)))
            ### Pass real data to netD
            netD.zero_grad()
            label = torch.full((b_size,), real_label, device=device)
            output = netD(realX).view(-1)
            errD_real = criterion(output, label)
            errD_real.backward()
            ### Pass fake data to netD
            noiseX = torch.randn(b_size, noise_dim, 1, 1, device=device)
            label.fill_(fake_label)
            fake = netG(noiseX)
            output = netD(fake.detach()).view(-1)
            errD_fake = criterion(output, label)
            errD_fake.backward()
            # Update D
            optimizerD.step()

            # Update G: maximize log(D(G(z)))
            netG.zero_grad()
            label.fill_(real_label)  # fake labels are real for generator
            output = netD(fake).view(-1)
            errG = criterion(output, label)
            errG.backward()
            # Update G
            optimizerG.step()

            ep_G_loss += errG.item()
            ep_D_loss += errD_real.item() + errD_fake.item()

        print(
            'epoch:', epoch, '/', num_epochs,
            '; G_loss:', ep_G_loss / (dataset.data.shape[0] // batch_size),
            '; D_loss:', ep_D_loss / (dataset.data.shape[0] // batch_size)
        )

        if epoch % 10 == 0 or (epoch+1) % num_epochs == 0:
            noiseX = torch.randn(5, noise_dim, 1, 1, device=device)
            fake = netG(noiseX)
            fig, ax = plt.subplots(1, 5)
            for i in range(5):
              ax[i].imshow((fake[i].detach().cpu().reshape(28, 28) + 1) * 0.5, cmap='gray', vmin=0, vmax=1)
              ax[i].axis('off')
            plt.show()
            plt.close()
            

Using device: cuda

/usr/local/lib/python3.7/dist-packages/torch/utils/data/dataloader.py:477: UserWarning: This DataLoader will create 4 worker processes in total. Our suggested max number of worker in current system is 2, which is smaller than what this DataLoader is going to create. Please be aware that excessive worker creation might get DataLoader running slow or even freeze, lower the worker number to avoid potential slowness/freeze if necessary.
  cpuset_checked))

epoch: 0 / 100 ; G_loss: 4.025594746837249 ; D_loss: 0.541045660198374

epoch: 1 / 100 ; G_loss: 2.2992326023741665 ; D_loss: 0.7424446938432052
epoch: 2 / 100 ; G_loss: 1.9783648646667473 ; D_loss: 0.8379886933069072
epoch: 3 / 100 ; G_loss: 1.9486756179577265 ; D_loss: 0.8742384373918812
epoch: 4 / 100 ; G_loss: 1.9037742032550085 ; D_loss: 0.9338736828798667
epoch: 5 / 100 ; G_loss: 1.8381108779173632 ; D_loss: 0.898899404077321
epoch: 6 / 100 ; G_loss: 1.9588675692331237 ; D_loss: 1.0968680807387319
epoch: 7 / 100 ; G_loss: 1.5689463105976071 ; D_loss: 0.9765682847390318
epoch: 8 / 100 ; G_loss: 1.7169391168997838 ; D_loss: 0.9516355206505356
epoch: 9 / 100 ; G_loss: 1.7341617144580581 ; D_loss: 0.9522137177678255
epoch: 10 / 100 ; G_loss: 1.7555983998836615 ; D_loss: 0.9518681125572095

epoch: 11 / 100 ; G_loss: 1.7360371795729694 ; D_loss: 0.957454572407863
epoch: 12 / 100 ; G_loss: 1.7592343311024528 ; D_loss: 0.9510215567663695
epoch: 13 / 100 ; G_loss: 1.7311208406901066 ; D_loss: 1.030192855028157
epoch: 14 / 100 ; G_loss: 1.7508108870595949 ; D_loss: 0.9457079759584024
epoch: 15 / 100 ; G_loss: 1.7789086168393111 ; D_loss: 0.942578107277807
epoch: 16 / 100 ; G_loss: 1.7989460865401814 ; D_loss: 0.9404596544674828
epoch: 17 / 100 ; G_loss: 1.78328772704316 ; D_loss: 0.9470169762006173
epoch: 18 / 100 ; G_loss: 1.829394889056173 ; D_loss: 0.9338661862425824
epoch: 19 / 100 ; G_loss: 1.8181682500021101 ; D_loss: 1.2435397097545777
epoch: 20 / 100 ; G_loss: 1.711103537143805 ; D_loss: 0.9201265703409146

epoch: 21 / 100 ; G_loss: 1.833623936288377 ; D_loss: 0.8992209309696132
epoch: 22 / 100 ; G_loss: 1.8796356265616214 ; D_loss: 0.8930723124271275
epoch: 23 / 100 ; G_loss: 1.8995193031608548 ; D_loss: 0.8997794571850035
epoch: 24 / 100 ; G_loss: 1.9316799252206445 ; D_loss: 0.8911645908672841
epoch: 25 / 100 ; G_loss: 1.941901578098281 ; D_loss: 0.8878339567563982
epoch: 26 / 100 ; G_loss: 1.9575555241770215 ; D_loss: 0.8912391072282424
epoch: 27 / 100 ; G_loss: 1.9775260340454233 ; D_loss: 0.8791944426285405
epoch: 28 / 100 ; G_loss: 1.9883574495712917 ; D_loss: 0.8786238700024083
epoch: 29 / 100 ; G_loss: 2.0078131208817163 ; D_loss: 0.8791541205798714
epoch: 30 / 100 ; G_loss: 2.0560114472849755 ; D_loss: 0.8606156882567283

epoch: 31 / 100 ; G_loss: 2.0541510882540646 ; D_loss: 0.863074517227773
epoch: 32 / 100 ; G_loss: 2.1322783022837637 ; D_loss: 1.2848561377587298
epoch: 33 / 100 ; G_loss: 1.9544953981525877 ; D_loss: 0.8356986431420869
epoch: 34 / 100 ; G_loss: 2.038284544634004 ; D_loss: 0.8263622351691254
epoch: 35 / 100 ; G_loss: 2.0833821836699786 ; D_loss: 0.8282250597850125
epoch: 36 / 100 ; G_loss: 2.0887566478843365 ; D_loss: 0.836319170287277
epoch: 37 / 100 ; G_loss: 2.1106385954170146 ; D_loss: 0.8263147123412699
epoch: 38 / 100 ; G_loss: 2.1214524405634303 ; D_loss: 0.8324364782112021
epoch: 39 / 100 ; G_loss: 2.1276926568940153 ; D_loss: 0.8409088224682033
epoch: 40 / 100 ; G_loss: 2.1308112964670882 ; D_loss: 0.8327076374911345

epoch: 41 / 100 ; G_loss: 2.131626808363148 ; D_loss: 0.8408964784003985
epoch: 42 / 100 ; G_loss: 2.1721376113147817 ; D_loss: 0.8258880305016397
epoch: 43 / 100 ; G_loss: 2.1578446550246997 ; D_loss: 0.8229683301069288
epoch: 44 / 100 ; G_loss: 2.177697014095437 ; D_loss: 0.8259326925294267
epoch: 45 / 100 ; G_loss: 2.194676972607262 ; D_loss: 0.8179371532402996
epoch: 46 / 100 ; G_loss: 1.8628630738816836 ; D_loss: 1.482671215695141
epoch: 47 / 100 ; G_loss: 1.9911826800586832 ; D_loss: 0.802754440655311
epoch: 48 / 100 ; G_loss: 2.1275273613695407 ; D_loss: 0.7862693232794603
epoch: 49 / 100 ; G_loss: 2.177294879132866 ; D_loss: 0.7863244345873339
epoch: 50 / 100 ; G_loss: 2.21672241606264 ; D_loss: 0.789539947691891

epoch: 51 / 100 ; G_loss: 2.2275218026250854 ; D_loss: 0.7933149426124799
epoch: 52 / 100 ; G_loss: 2.22218901428402 ; D_loss: 0.8077221612294769
epoch: 53 / 100 ; G_loss: 2.2164184894317236 ; D_loss: 0.8100897765153239
epoch: 54 / 100 ; G_loss: 2.2335894105271397 ; D_loss: 0.7955706094065284
epoch: 55 / 100 ; G_loss: 2.240376371476385 ; D_loss: 0.8128903179915032
epoch: 56 / 100 ; G_loss: 2.234919564336793 ; D_loss: 0.8055409416237957
epoch: 57 / 100 ; G_loss: 2.2421333145382056 ; D_loss: 0.8070204915462905
epoch: 58 / 100 ; G_loss: 2.2626537757042127 ; D_loss: 0.8084046615558302
epoch: 59 / 100 ; G_loss: 2.2467449298526487 ; D_loss: 0.8024482933533752
epoch: 60 / 100 ; G_loss: 2.026891887698301 ; D_loss: 2.1036019780422976

epoch: 61 / 100 ; G_loss: 1.9626058108276792 ; D_loss: 0.797756417757935
epoch: 62 / 100 ; G_loss: 2.1288901776330085 ; D_loss: 0.7584441274913967
epoch: 63 / 100 ; G_loss: 2.2040701368425646 ; D_loss: 0.7601017936363689
epoch: 64 / 100 ; G_loss: 2.2573280049185467 ; D_loss: 0.7619534540189128
epoch: 65 / 100 ; G_loss: 2.2657612470480113 ; D_loss: 0.7767149130375977
epoch: 66 / 100 ; G_loss: 2.2908432669619208 ; D_loss: 0.7709830653750234
epoch: 67 / 100 ; G_loss: 2.2890092284760923 ; D_loss: 0.7892235971223085
epoch: 68 / 100 ; G_loss: 2.2841589216493134 ; D_loss: 0.795194176877411
epoch: 69 / 100 ; G_loss: 2.294537895000898 ; D_loss: 0.7916370539679232
epoch: 70 / 100 ; G_loss: 2.2995976207093296 ; D_loss: 0.7932863374296416

epoch: 71 / 100 ; G_loss: 2.2877162328133216 ; D_loss: 0.8016409057423345
epoch: 72 / 100 ; G_loss: 2.2639855719529667 ; D_loss: 0.8041345239258729
epoch: 73 / 100 ; G_loss: 2.2871644779657707 ; D_loss: 0.7964961169621884
epoch: 74 / 100 ; G_loss: 2.298080288956308 ; D_loss: 0.7952589884591408
epoch: 75 / 100 ; G_loss: 2.3144368096294565 ; D_loss: 0.7932604627094717
epoch: 76 / 100 ; G_loss: 2.319773532386519 ; D_loss: 0.7938909871010189
epoch: 77 / 100 ; G_loss: 2.336074157148345 ; D_loss: 0.7911147294072514
epoch: 78 / 100 ; G_loss: 2.8829208509505326 ; D_loss: 2.608052038170312
epoch: 79 / 100 ; G_loss: 1.8106857831152077 ; D_loss: 0.8609801237909203
epoch: 80 / 100 ; G_loss: 2.0663085607891407 ; D_loss: 0.7559228680518448

epoch: 81 / 100 ; G_loss: 2.1988770847137156 ; D_loss: 0.7310278601307645
epoch: 82 / 100 ; G_loss: 2.28019866372785 ; D_loss: 0.7315544899966981
epoch: 83 / 100 ; G_loss: 2.3286844176104946 ; D_loss: 0.7390283906243296
epoch: 84 / 100 ; G_loss: 2.3573762165684986 ; D_loss: 0.7349227956918061
epoch: 85 / 100 ; G_loss: 2.3894227648902144 ; D_loss: 0.747650688338993
epoch: 86 / 100 ; G_loss: 2.3921624722643795 ; D_loss: 0.7551322772653184
epoch: 87 / 100 ; G_loss: 2.41464268295174 ; D_loss: 0.7602824414801649
epoch: 88 / 100 ; G_loss: 2.4098722081408543 ; D_loss: 0.7749648282511367
epoch: 89 / 100 ; G_loss: 2.4061698116298413 ; D_loss: 0.7687831522785445
epoch: 90 / 100 ; G_loss: 2.4115507567030754 ; D_loss: 0.7742021458271222

epoch: 91 / 100 ; G_loss: 2.409960554451005 ; D_loss: 0.7709544651114788
epoch: 92 / 100 ; G_loss: 2.4242513600068216 ; D_loss: 0.7755984421341847
epoch: 93 / 100 ; G_loss: 2.1453172233505886 ; D_loss: 1.5187836557232048
epoch: 94 / 100 ; G_loss: 2.2327247835122623 ; D_loss: 0.7408213297971803
epoch: 95 / 100 ; G_loss: 2.3568259867338033 ; D_loss: 0.7247299123396221
epoch: 96 / 100 ; G_loss: 2.425844599039127 ; D_loss: 0.7328249701163453
epoch: 97 / 100 ; G_loss: 2.4568663583352017 ; D_loss: 0.7322179988090299
epoch: 98 / 100 ; G_loss: 2.4941434651358514 ; D_loss: 0.7371676288959053
epoch: 99 / 100 ; G_loss: 2.488156497478485 ; D_loss: 0.7438272241120919