\n Generative Adversarial Networks (GAN) is one of the fundamental deep learning models proposed by Ian Goodfellow in 2014. GAN consists of two neural networks: \n The structure of GAN operates as follows:\n <\/p>\n \n The training process of GAN consists of the following iterative steps:\n <\/p>\n \n Now, let’s implement GAN in PyTorch. Our goal is to generate digit images using the MNIST dataset.\n <\/p>\n \n The MNIST dataset can be easily retrieved through PyTorch’s torchvision library. The code below shows how to load and preprocess the data.\n <\/p>\n \n Now we define the two networks of GAN. We will create the generator and discriminator using a simple neural network structure.\n <\/p>\n \n The loss function used for both the generator and discriminator is binary cross-entropy loss. We adopt the Adam optimization algorithm.\n <\/p>\n \n Training proceeds as follows:\n <\/p>\n \n After training is complete, we visualize the images generated.\n <\/p>\n \n GANs can be used in various fields beyond image generation. For example, they are employed in style transfer, image restoration, video generation, and have garnered significant attention in the field of artificial intelligence. \n In this tutorial, we learned the basics of implementing GANs using PyTorch and understood how GAN operates through actual code. GAN is a technology that will continue to evolve and holds great potential in various applications. <\/body><\/p>\n","protected":false},"excerpt":{"rendered":"
\n Generator<\/strong> and Discriminator<\/strong>. The generator tries to create fake data, while the discriminator attempts to determine whether the data is real or fake.
\n These two networks compete against each other during training, hence the term “Adversarial.” This process is unsupervised, allowing the generator to produce data that increasingly resembles real data.\n <\/p>\n2. Structure of GAN<\/h2>\n
\n
3. How GAN Works<\/h2>\n
\n
4. Implementing GAN in PyTorch<\/h2>\n
4.1 Installing Required Libraries<\/h3>\n
\n
pip install torch torchvision matplotlib<\/code>\n <\/pre>\n4.2 Preparing the Dataset<\/h3>\n
\n
\nimport torch\nfrom torchvision import datasets, transforms\nfrom torch.utils.data import DataLoader\n\n# Data transformation settings\ntransform = transforms.Compose([\n transforms.ToTensor(),\n transforms.Normalize((0.5,), (0.5,))\n])\n\n# Load MNIST dataset\ntrain_dataset = datasets.MNIST(root='.\/data', train=True, download=True, transform=transform)\ntrain_loader = DataLoader(train_dataset, batch_size=64, shuffle=True)\n <\/code>\n <\/pre>\n4.3 Defining the Generator and Discriminator Networks<\/h3>\n
\n
\nimport torch.nn as nn\n\n# Define Generator Network\nclass Generator(nn.Module):\n def __init__(self):\n super(Generator, self).__init__()\n self.model = nn.Sequential(\n nn.Linear(100, 256),\n nn.ReLU(),\n nn.Linear(256, 512),\n nn.ReLU(),\n nn.Linear(512, 1024),\n nn.ReLU(),\n nn.Linear(1024, 28 * 28),\n nn.Tanh()\n )\n\n def forward(self, z):\n return self.model(z).view(-1, 1, 28, 28)\n\n# Define Discriminator Network\nclass Discriminator(nn.Module):\n def __init__(self):\n super(Discriminator, self).__init__()\n self.model = nn.Sequential(\n nn.Flatten(),\n nn.Linear(28 * 28, 512),\n nn.LeakyReLU(0.2),\n nn.Linear(512, 256),\n nn.LeakyReLU(0.2),\n nn.Linear(256, 1),\n nn.Sigmoid()\n )\n\n def forward(self, img):\n return self.model(img)\n <\/code>\n <\/pre>\n4.4 Setting Loss Function and Optimization Algorithm<\/h3>\n
\n
\nimport torch.optim as optim\n\n# Define loss function and optimization algorithm\ncriterion = nn.BCELoss()\ngenerator = Generator()\ndiscriminator = Discriminator()\noptimizer_G = optim.Adam(generator.parameters(), lr=0.0002, betas=(0.5, 0.999))\noptimizer_D = optim.Adam(discriminator.parameters(), lr=0.0002, betas=(0.5, 0.999))\n <\/code>\n <\/pre>\n4.5 Implementing the Training Process<\/h3>\n
\n
\nimport numpy as np\n\nnum_epochs = 50\nfor epoch in range(num_epochs):\n for i, (images, _) in enumerate(train_loader):\n batch_size = images.size(0)\n \n # Set real and fake labels\n real_labels = torch.ones(batch_size, 1)\n fake_labels = torch.zeros(batch_size, 1)\n\n # Train Discriminator\n outputs = discriminator(images)\n d_loss_real = criterion(outputs, real_labels)\n\n noise = torch.randn(batch_size, 100)\n fake_images = generator(noise)\n outputs = discriminator(fake_images.detach())\n d_loss_fake = criterion(outputs, fake_labels)\n\n d_loss = d_loss_real + d_loss_fake\n optimizer_D.zero_grad()\n d_loss.backward()\n optimizer_D.step()\n\n # Train Generator\n outputs = discriminator(fake_images)\n g_loss = criterion(outputs, real_labels)\n\n optimizer_G.zero_grad()\n g_loss.backward()\n optimizer_G.step()\n\n # Output training status\n if (i + 1) % 100 == 0:\n print(f'Epoch [{epoch + 1}\/{num_epochs}], Step [{i + 1}\/{len(train_loader)}], d_loss: {d_loss.item():.4f}, g_loss: {g_loss.item():.4f}')\n <\/code>\n <\/pre>\n4.6 Visualizing Generated Images<\/h3>\n
\n
\nimport matplotlib.pyplot as plt\n\ndef visualize_generator(num_images):\n noise = torch.randn(num_images, 100)\n with torch.no_grad():\n generated_images = generator(noise)\n\n plt.figure(figsize=(10, 10))\n for i in range(num_images):\n plt.subplot(5, 5, i + 1)\n plt.imshow(generated_images[i][0].cpu().numpy(), cmap='gray')\n plt.axis('off')\n plt.show()\n\nvisualize_generator(25)\n <\/code>\n <\/pre>\n5. Applications of GAN<\/h2>\n
\n The development of GANs is revealing new possibilities through generative models.\n <\/p>\n6. Conclusion<\/h2>\n
\n I recommend exploring various modified models based on GAN in the future.\n <\/p>\n