GAN(Generative Adversarial Networks)\uc740 2014\ub144 Ian Goodfellow\uc5d0 \uc758\ud574 \uc81c\uc548\ub41c \ud601\uc2e0\uc801\uc778 \ub525\ub7ec\ub2dd \ubaa8\ub378\ub85c, \uc0dd\uc131 \ubaa8\ub378\uacfc \ud310\ubcc4 \ubaa8\ub378\uc774 \uc11c\ub85c \ub300\ub9bd\ud558\uc5ec \ud559\uc2b5\ud558\ub294 \ubc29\uc2dd\uc73c\ub85c \uc791\ub3d9\ud569\ub2c8\ub2e4. GAN\uc758 \uae30\ubcf8 \uad6c\uc131 \uc694\uc18c\ub294 \uc0dd\uc131\uae30(Generator)\uc640 \ud310\ubcc4\uae30(Discriminator)\uc785\ub2c8\ub2e4. \uc0dd\uc131\uae30\ub294 \uc2e4\uc81c\uc640 \uc720\uc0ac\ud55c \ub370\uc774\ud130\ub97c \uc0dd\uc131\ud558\ub824\uace0 \ud558\uba70, \ud310\ubcc4\uae30\ub294 \uc8fc\uc5b4\uc9c4 \ub370\uc774\ud130\uac00 \uc2e4\uc81c \ub370\uc774\ud130\uc778\uc9c0 \uc0dd\uc131\ub41c \ub370\uc774\ud130\uc778\uc9c0\ub97c \ud310\ub2e8\ud569\ub2c8\ub2e4. \uc774\ub7ec\ud55c \ub450 \ubaa8\ub378 \uac04\uc758 \uacbd\uc7c1\uc744 \ud1b5\ud574 \uc0dd\uc131\uae30\ub294 \uc810\uc810 \ub354 \uc2e4\uc81c\uc640 \uc720\uc0ac\ud55c \ub370\uc774\ud130\ub97c \ub9cc\ub4e4\uc5b4 \ub0b4\uac8c \ub429\ub2c8\ub2e4.<\/p>\n
\uc0dd\uc131\uae30\ub294 \ub79c\ub364 \ub178\uc774\uc988 \ubca1\ud130\ub97c \uc785\ub825\ubc1b\uc544 \uc2e4\uc81c \ub370\uc774\ud130\uc640 \uc720\uc0ac\ud55c \ub370\uc774\ud130\ub97c \uc0dd\uc131\ud558\ub294 \uc2e0\uacbd\ub9dd\uc785\ub2c8\ub2e4. \uc774 \ub124\ud2b8\uc6cc\ud06c\ub294 \uc77c\ubc18\uc801\uc73c\ub85c \ub2e4\uce35 \ud37c\uc149\ud2b8\ub860 \ub610\ub294 \ucee8\ubcfc\ub8e8\uc158 \uc2e0\uacbd\ub9dd\uc744 \uc0ac\uc6a9\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
\ud310\ubcc4\uae30\ub294 \uc785\ub825\ubc1b\uc740 \ub370\uc774\ud130\uac00 \uc2e4\uc81c \ub370\uc774\ud130\uc778\uc9c0 \uc0dd\uc131\ub41c \ub370\uc774\ud130\uc778\uc9c0\ub97c \ud310\ub2e8\ud558\ub294 \uc2e0\uacbd\ub9dd\uc785\ub2c8\ub2e4. \uc0dd\uc131\uae30\uac00 \ub9cc\ub4e0 \uac00\uc9dc \ub370\uc774\ud130\uc640 \uc2e4\uc81c \ub370\uc774\ud130\ub97c \uc798 \uad6c\ubd84\ud560 \uc218 \uc788\ub294 \ub2a5\ub825\uc774 \uc911\uc694\ud569\ub2c8\ub2e4.<\/p>\n
GAN\uc758 \uc190\uc2e4 \ud568\uc218\ub294 \uc0dd\uc131\uae30\uc758 \uc190\uc2e4\uacfc \ud310\ubcc4\uae30\uc758 \uc190\uc2e4\ub85c \ub098\ub269\ub2c8\ub2e4. \uc0dd\uc131\uae30\uc758 \ubaa9\ud45c\ub294 \ud310\ubcc4\uae30\ub97c \uc18d\uc774\ub294 \uac83\uc774\uace0, \ud310\ubcc4\uae30\uc758 \ubaa9\ud45c\ub294 \uc0dd\uc131\ub41c \ub370\uc774\ud130\ub97c \uc798 \uad6c\ubcc4\ud558\ub294 \uac83\uc785\ub2c8\ub2e4.<\/p>\n
GAN\uc758 \ud6c8\ub828 \uacfc\uc815\uc740 \ubc18\ubcf5\uc801\uc774\uba70, \ub2e4\uc74c\uacfc \uac19\uc740 \ub2e8\uacc4\ub85c \uad6c\uc131\ub429\ub2c8\ub2e4:<\/p>\n
GAN\uc740 \ub2e4\uc74c\uacfc \uac19\uc740 \ub2e4\uc591\ud55c \ubd84\uc57c\uc5d0\uc11c \ud65c\uc6a9\ub429\ub2c8\ub2e4:<\/p>\n
\uc815\ud615 \ub370\uc774\ud130\ub294 \uad6c\uc870\ud654\ub41c \ub370\uc774\ud130\ub85c, \uad00\uacc4\ud615 \ub370\uc774\ud130\ubca0\uc774\uc2a4\uc5d0 \uc27d\uac8c \ud45c\ud604\ud560 \uc218 \uc788\ub294 \ub370\uc774\ud130\uc785\ub2c8\ub2e4. \ubc18\uba74, \ube44\uc815\ud615 \ub370\uc774\ud130\ub294 \ud14d\uc2a4\ud2b8, \uc774\ubbf8\uc9c0, \ube44\ub514\uc624 \ub4f1\uacfc \uac19\uc774 \uace0\uc720\ud55c \ud615\uc2dd\uacfc \uad6c\uc870\uac00 \uc5c6\ub294 \ub370\uc774\ud130\uc785\ub2c8\ub2e4. GAN\uc740 \ube44\uc815\ud615 \ub370\uc774\ud130\uc5d0 \uc8fc\ub85c \uc0ac\uc6a9\ub418\uc9c0\ub9cc \uc815\ud615 \ub370\uc774\ud130\uc5d0\ub3c4 \ud65c\uc6a9\ub420 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
\ub2e4\uc74c\uc740 \ud30c\uc774\ud1a0\uce58\ub97c \uc0ac\uc6a9\ud55c GAN\uc758 \uac04\ub2e8\ud55c \uad6c\ud604 \uc608\uc81c\uc785\ub2c8\ub2e4. \uc774 \uc608\uc81c\uc5d0\uc11c\ub294 MNIST \ub370\uc774\ud130\uc14b\uc744 \uc0ac\uc6a9\ud558\uc5ec \uc190\uae00\uc528 \uc22b\uc790\ub97c \uc0dd\uc131\ud569\ub2c8\ub2e4.<\/p>\n
\uba3c\uc800 \ud544\uc694\ud55c \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \uc124\uce58\ud558\uace0 \ubd88\ub7ec\uc635\ub2c8\ub2e4.<\/p>\n
!pip install torch torchvision matplotlib<\/code><\/pre>\n6.2 \ub370\uc774\ud130\uc14b \uc900\ube44<\/h3>\n
MNIST \ub370\uc774\ud130\uc14b\uc744 \ub85c\ub4dc\ud558\uace0, \uc774\ub97c \ud150\uc11c\ub85c \ubcc0\ud658\ud569\ub2c8\ub2e4.<\/p>\n
import torch\nimport torchvision\nimport torchvision.transforms as transforms\n\ntransform = transforms.Compose([\n transforms.ToTensor(),\n transforms.Normalize((0.5,), (0.5,)),\n])\n\ntrain_dataset = torchvision.datasets.MNIST(root='.\/data', train=True, download=True, transform=transform)\ntrain_loader = torch.utils.data.DataLoader(dataset=train_dataset, batch_size=64, shuffle=True)<\/code><\/pre>\n6.3 \ubaa8\ub378 \uad6c\uc131<\/h3>\n
\uc0dd\uc131\uae30\uc640 \ud310\ubcc4\uae30 \ubaa8\ub378\uc744 \uc815\uc758\ud569\ub2c8\ub2e4.<\/p>\n
import torch.nn as nn\n\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, 784), # MNIST\uc640 \uac19\uc740 28x28 \uc774\ubbf8\uc9c0\n nn.Tanh(),\n )\n\n def forward(self, x):\n return self.model(x)\n\nclass Discriminator(nn.Module):\n def __init__(self):\n super(Discriminator, self).__init__()\n self.model = nn.Sequential(\n nn.Linear(784, 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, x):\n return self.model(x)<\/code><\/pre>\n6.4 \uc190\uc2e4 \ud568\uc218\uc640 \uc635\ud2f0\ub9c8\uc774\uc800 \uc124\uc815<\/h3>\n
\uc190\uc2e4 \ud568\uc218\ub85c\ub294 BCELoss\ub97c \uc0ac\uc6a9\ud558\uace0, Adam \uc635\ud2f0\ub9c8\uc774\uc800\ub85c \ud559\uc2b5\uc744 \uc9c4\ud589\ud569\ub2c8\ub2e4.<\/p>\n
criterion = nn.BCELoss()\ngenerator = Generator()\ndiscriminator = Discriminator()\n\noptimizer_G = torch.optim.Adam(generator.parameters(), lr=0.0002, betas=(0.5, 0.999))\noptimizer_D = torch.optim.Adam(discriminator.parameters(), lr=0.0002, betas=(0.5, 0.999))<\/code><\/pre>\n6.5 \ubaa8\ub378 \ud6c8\ub828<\/h3>\n
\ubaa8\ub378 \ud6c8\ub828\uc744 \uc218\ud589\ud569\ub2c8\ub2e4. \ub9e4 \uc5d0\ud3ec\ud06c\ub9c8\ub2e4 \uc0dd\uc131\ub41c \uc0d8\ud50c\uc744 \ud655\uc778\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
import matplotlib.pyplot as plt\n\nnum_epochs = 100\nfor epoch in range(num_epochs):\n for i, (real_images, _) in enumerate(train_loader):\n batch_size = real_images.size(0)\n \n # \ub77c\ubca8 \uc0dd\uc131\n real_labels = torch.ones(batch_size, 1)\n fake_labels = torch.zeros(batch_size, 1)\n\n # \ud310\ubcc4\uae30 \ud6c8\ub828\n optimizer_D.zero_grad()\n outputs = discriminator(real_images.view(batch_size, -1))\n d_loss_real = criterion(outputs, real_labels)\n d_loss_real.backward()\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 d_loss_fake.backward()\n optimizer_D.step()\n\n # \uc0dd\uc131\uae30 \ud6c8\ub828\n optimizer_G.zero_grad()\n outputs = discriminator(fake_images)\n g_loss = criterion(outputs, real_labels)\n g_loss.backward()\n optimizer_G.step()\n\n if (epoch+1) % 10 == 0:\n print(f'Epoch [{epoch+1}\/{num_epochs}], d_loss: {d_loss_real.item() + d_loss_fake.item():.4f}, g_loss: {g_loss.item():.4f}')\n # \uc0dd\uc131\ub41c \uc774\ubbf8\uc9c0 \ud655\uc778\n with torch.no_grad():\n fake_images = generator(noise).view(-1, 1, 28, 28)\n grid = torchvision.utils.make_grid(fake_images, nrow=8, normalize=True)\n plt.imshow(grid.permute(1, 2, 0).numpy(), cmap='gray')\n plt.show()<\/code><\/pre>\n6.6 \uacb0\uacfc \ud655\uc778<\/h3>\n
\ud6c8\ub828\uc774 \uc9c4\ud589\ub428\uc5d0 \ub530\ub77c \uc0dd\uc131\uae30\ub294 \uc810\uc810 \ub354 \ud604\uc2e4\uc801\uc778 \uc190\uae00\uc528 \uc22b\uc790\ub97c \uc0dd\uc131\ud569\ub2c8\ub2e4. \uc774\ub97c \ud1b5\ud574 GAN\uc758 \uc131\ub2a5\uc744 \ud655\uc778\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
7. \uc885\ub8cc \ubc0f \uacb0\ub860<\/h2>\n
\uc774\ubc88 \uae00\uc5d0\uc11c\ub294 GAN\uc758 \uac1c\uc694, \uad6c\uc131 \uc694\uc18c, \ud6c8\ub828 \uacfc\uc815, \ud30c\uc774\ud1a0\uce58\ub97c \uc774\uc6a9\ud55c \uad6c\ud604 \ubc29\ubc95\uc744 \ub2e4\ub8e8\uc5c8\uc2b5\ub2c8\ub2e4. GAN\uc740 \ube44\uc815\ud615 \ub370\uc774\ud130 \uc0dd\uc131\uc5d0\uc11c \ub6f0\uc5b4\ub09c \uc131\ub2a5\uc744 \ubcf4\uc5ec\uc8fc\uba70, \ub2e4\uc591\ud55c \uc751\uc6a9 \ubd84\uc57c\uc5d0\uc11c \ud65c\uc6a9\ub420 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc774\ub7ec\ud55c \uae30\uc220\uc774 \ubc1c\uc804\ud568\uc5d0 \ub530\ub77c \uc55e\uc73c\ub85c\uc758 \uac00\ub2a5\uc131\uc740 \ubb34\uad81\ubb34\uc9c4\ud569\ub2c8\ub2e4.<\/p>\n
8. \ucc38\uace0 \ubb38\ud5cc<\/h2>\n
1. Ian Goodfellow et al., “Generative Adversarial Networks”, NeurIPS 2014.<\/p>\n
2. PyTorch Documentation – https:\/\/pytorch.org\/docs\/stable\/index.html<\/a><\/p>\n3. torchvision Documentation – https:\/\/pytorch.org\/vision\/stable\/index.html<\/a><\/p>\n<\/body><\/p>\n","protected":false},"excerpt":{"rendered":"