\ub525\ub7ec\ub2dd\uc758 \ubc1c\uc804\uacfc \ud568\uaed8 GAN(Generative Adversarial Network)\uc774\ub77c\ub294 \ud504\ub808\uc784\uc6cc\ud06c\ub294 \uc774\ubbf8\uc9c0 \uc0dd\uc131, \ubcc0\ud658 \ubc0f \ud3b8\uc9d1 \ubd84\uc57c\uc5d0\uc11c \ud601\uc2e0\uc801\uc778 \ubcc0\ud654\ub97c \uac00\uc838\uc654\uc2b5\ub2c8\ub2e4. GAN\uc740 \ub450 \uac1c\uc758 \uc2e0\uacbd\ub9dd\uc778 \uc0dd\uc131\uc790(Generator)\uc640 \ud310\ubcc4\uc790(Discriminator)\uac00 \uc11c\ub85c \uacbd\uc7c1\ud558\uba70 \ud559\uc2b5\ud558\ub294 \uad6c\uc870\ub85c \ub418\uc5b4 \uc788\uc2b5\ub2c8\ub2e4. \uc774 \uae00\uc5d0\uc11c\ub294 GAN\uc758 \uae30\ubcf8 \uac1c\ub150, \uc791\ub3d9 \uc6d0\ub9ac, \ud30c\uc774\ud1a0\uce58(PyTorch)\ub97c \uc774\uc6a9\ud55c \uad6c\ud604 \uc608\uc2dc, \uadf8\ub9ac\uace0 GAN\uc744 \ud65c\uc6a9\ud55c \uc774\ubbf8\uc9c0 \uc0dd\uc131\uc758 \ubc1c\uc804\uc5d0 \ub300\ud574 \uc2ec\ub3c4 \uc788\uac8c \ub2e4\ub8e8\uc5b4 \ubcf4\uaca0\uc2b5\ub2c8\ub2e4.<\/p>\n
GAN\uc740 Ian Goodfellow \ub4f1\uc774 2014\ub144\uc5d0 \uc81c\uc548\ud55c \ubaa8\ub378\ub85c, \ub450 \uac1c\uc758 \uc2e0\uacbd\ub9dd\uc774 \uc801\ub300\uc801\uc778 \uad00\uacc4\ub97c \ud1b5\ud574 \ud559\uc2b5\uc744 \uc9c4\ud589\ud569\ub2c8\ub2e4. \uc0dd\uc131\uc790\ub294 \uac00\uc9dc \uc774\ubbf8\uc9c0\ub97c \uc0dd\uc131\ud558\uace0, \ud310\ubcc4\uc790\ub294 \uc9c4\uc9dc \uc774\ubbf8\uc9c0\uc640 \uac00\uc9dc \uc774\ubbf8\uc9c0\ub97c \uad6c\ubcc4\ud558\ub294 \uc5ed\ud560\uc744 \ud569\ub2c8\ub2e4. \uc774 \uacfc\uc815\uc740 \ub2e4\uc74c\uacfc \uac19\uc774 \uc9c4\ud589\ub429\ub2c8\ub2e4:<\/p>\n
GAN\uc758 \uae30\ubcf8 \uad6c\uc131 \uc694\uc18c\ub294 \uc0dd\uc131\uc790\uc640 \ud310\ubcc4\uc790\ub85c, \uc774\ub4e4\uc740 \ub2e4\uc74c\uacfc \uac19\uc740 \uc5ed\ud560\uc744 \uc218\ud589\ud569\ub2c8\ub2e4:<\/p>\n
GAN\uc758 \uc190\uc2e4 \ud568\uc218\ub294 \ub2e4\uc74c\uacfc \uac19\uc774 \uc815\uc758\ub429\ub2c8\ub2e4:<\/p>\n
\ud310\ubcc4\uc790\uc758 \uc190\uc2e4 \ud568\uc218\ub294 \uc9c4\uc9dc \uc774\ubbf8\uc9c0\uc640 \uac00\uc9dc \uc774\ubbf8\uc9c0\uc5d0 \ub300\ud55c \uc608\uce21\uc744 \ucd5c\ub300\ud654\ud558\ub824\uace0 \ud569\ub2c8\ub2e4.<\/p>\n
\nD_loss = -E[log(D(x))] - E[log(1 - D(G(z)))]\n<\/pre>\n\uc0dd\uc131\uc790\uc758 \uc190\uc2e4 \ud568\uc218\ub294 \ud310\ubcc4\uc790\uac00 \uac00\uc9dc \uc774\ubbf8\uc9c0\ub97c \uc9c4\uc9dc\ub85c \uc798\ubabb \ubd84\ub958\ud558\ub3c4\ub85d \ud559\uc2b5\ud569\ub2c8\ub2e4.<\/p>\n
\nG_loss = -E[log(D(G(z)))]\n<\/pre>\n2. \ud30c\uc774\ud1a0\uce58\ub97c \uc774\uc6a9\ud55c GAN \uad6c\ud604<\/h2>\n
\uc774\uc81c \ud30c\uc774\ud1a0\uce58\ub97c \uc0ac\uc6a9\ud558\uc5ec \uac04\ub2e8\ud55c GAN\uc744 \uad6c\ud604\ud574 \ubcf4\uaca0\uc2b5\ub2c8\ub2e4. \uc774\ub97c \ud1b5\ud574 GAN\uc758 \uc791\ub3d9 \ubc29\uc2dd\uc744 \uc2e4\uc2b5\ud558\uace0, \uc774\ubbf8\uc9c0\ub97c \uc0dd\uc131\ud558\ub294 \uacfc\uc815\uc744 \uc2dc\uac01\uc801\uc73c\ub85c \uc774\ud574\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
2.1 \ud544\uc694\ud55c \ub77c\uc774\ube0c\ub7ec\ub9ac \uc124\uce58<\/h3>\n
\ud30c\uc774\ud1a0\uce58 \ubc0f torchvision\uc744 \uc124\uce58\ud569\ub2c8\ub2e4. \uc774\uac83\uc740 \uc2e0\uacbd\ub9dd \uad6c\ucd95\uacfc \ub370\uc774\ud130\uc14b \ub85c\ub529\uc5d0 \ud544\uc694\ud569\ub2c8\ub2e4.<\/p>\n
\npip install torch torchvision\n<\/pre>\n2.2 \ub370\uc774\ud130\uc14b \uc900\ube44<\/h3>\n
MNIST \ub370\uc774\ud130\uc14b\uc744 \uc0ac\uc6a9\ud558\uc5ec \uc22b\uc790 \uc774\ubbf8\uc9c0\ub97c \uc0dd\uc131\ud574\ubd05\ub2c8\ub2e4.<\/p>\n
\nimport torch\nimport torchvision\nimport torchvision.transforms as transforms\n\n# MNIST \ub370\uc774\ud130\uc14b \ub85c\ub4dc\ntransform = transforms.Compose([transforms.ToTensor(), transforms.Normalize((0.5,), (0.5,))])\ntrainset = torchvision.datasets.MNIST(root='.\/data', train=True, download=True, transform=transform)\ntrainloader = torch.utils.data.DataLoader(trainset, batch_size=64, shuffle=True)\n<\/pre>\n2.3 \uc0dd\uc131\uc790 \ubc0f \ud310\ubcc4\uc790 \ubaa8\ub378 \uc815\uc758<\/h3>\n
\nimport torch.nn as nn\n\n# \uc0dd\uc131\uc790 \uc815\uc758\nclass Generator(nn.Module):\n def __init__(self):\n super(Generator, self).__init__()\n self.fc = 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.fc(z).view(-1, 1, 28, 28)\n\n# \ud310\ubcc4\uc790 \uc815\uc758\nclass Discriminator(nn.Module):\n def __init__(self):\n super(Discriminator, self).__init__()\n self.fc = nn.Sequential(\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, x):\n return self.fc(x.view(-1, 28 * 28))\n<\/pre>\n2.4 \uc190\uc2e4 \ud568\uc218 \ubc0f \ucd5c\uc801\ud654\uae30 \uc124\uc815<\/h3>\n
\nimport torch.optim as optim\n\n# \ubaa8\ub378 \uc778\uc2a4\ud134\uc2a4 \uc0dd\uc131\nG = Generator()\nD = Discriminator()\n\n# \uc190\uc2e4 \ud568\uc218 \ubc0f \ucd5c\uc801\ud654\uae30 \uc124\uc815\ncriterion = nn.BCELoss()\noptimizer_G = optim.Adam(G.parameters(), lr=0.0002, betas=(0.5, 0.999))\noptimizer_D = optim.Adam(D.parameters(), lr=0.0002, betas=(0.5, 0.999))\n<\/pre>\n2.5 GAN \ud559\uc2b5 \ub8e8\ud504<\/h3>\n
\uc774\uc81c GAN\uc744 \ud559\uc2b5\uc2dc\ud0ac \ucc28\ub840\uc785\ub2c8\ub2e4. \ud559\uc2b5 \ubc18\ubcf5\ubb38\uc744 \ud1b5\ud574 \uc0dd\uc131\uc790\uc640 \ud310\ubcc4\uc790\ub97c \uad50\ub300\ub85c \uc5c5\ub370\uc774\ud2b8\ud569\ub2c8\ub2e4.<\/p>\n
\nnum_epochs = 200\nfor epoch in range(num_epochs):\n for i, (real_images, _) in enumerate(trainloader):\n # \uc9c4\uc9dc\uc640 \uac00\uc9dc\ub97c \uc704\ud55c \ub808\uc774\ube14 \ub9cc\ub4e4\uae30\n real_labels = torch.ones(real_images.size(0), 1)\n fake_labels = torch.zeros(real_images.size(0), 1)\n\n # \ud310\ubcc4\uc790 \ud559\uc2b5\n optimizer_D.zero_grad()\n outputs = D(real_images)\n d_loss_real = criterion(outputs, real_labels)\n\n z = torch.randn(real_images.size(0), 100)\n fake_images = G(z)\n outputs = D(fake_images.detach())\n d_loss_fake = criterion(outputs, fake_labels)\n\n d_loss = d_loss_real + d_loss_fake\n d_loss.backward()\n optimizer_D.step()\n\n # \uc0dd\uc131\uc790 \ud559\uc2b5\n optimizer_G.zero_grad()\n outputs = D(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.item():.4f}, g_loss: {g_loss.item():.4f}')\n<\/pre>\n2.6 \uc774\ubbf8\uc9c0 \uc0dd\uc131 \uc2dc\uac01\ud654<\/h3>\n
\ud559\uc2b5\uc774 \uc644\ub8cc\ub41c \ud6c4\uc5d0\ub294 \uc0dd\uc131\uc790\ub97c \uc0ac\uc6a9\ud558\uc5ec \uc774\ubbf8\uc9c0\ub97c \uc0dd\uc131\ud558\uace0 \uc2dc\uac01\ud654\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
\nimport matplotlib.pyplot as plt\nimport numpy as np\n\nz = torch.randn(64, 100)\nfake_images = G(z)\n\n# \uc0dd\uc131\ub41c \uc774\ubbf8\uc9c0 \uc2dc\uac01\ud654\ngrid = torchvision.utils.make_grid(fake_images, nrow=8, normalize=True)\nplt.imshow(np.transpose(grid.detach().numpy(), (1, 2, 0)))\nplt.axis('off')\nplt.show()\n<\/pre>\n3. GAN\uc758 \ubc1c\uc804\uacfc \uc751\uc6a9<\/h2>\n
GAN\uc740 \uc774\ubbf8\uc9c0\ub97c \uc0dd\uc131\ud558\ub294 \uac83 \uc678\uc5d0\ub3c4 \ub2e4\uc591\ud55c \ubd84\uc57c\uc5d0\uc11c \ud65c\uc6a9\ub418\uace0 \uc788\uc2b5\ub2c8\ub2e4. \uc608\ub97c \ub4e4\uc5b4:<\/p>\n
\n
- \uc2a4\ud0c0\uc77c \ubcc0\ud658<\/strong>: \uc0ac\uc9c4\uc758 \uc2a4\ud0c0\uc77c\uc744 \ub2e4\ub978 \uc2a4\ud0c0\uc77c\ub85c \ubcc0\ud658\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/li>\n
- \uc774\ubbf8\uc9c0 \ubcf4\uc644<\/strong>: \uacb0\uc190\ub41c \uc774\ubbf8\uc9c0 \ubd80\ubd84\uc744 \uc0dd\uc131\ud558\uc5ec \uc644\uc804\ud55c \uc774\ubbf8\uc9c0\ub97c \ubcf5\uc6d0\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/li>\n
- \uc288\ud37c \ud574\uc0c1\ub3c4<\/strong>: \uc800\ud574\uc0c1\ub3c4 \uc774\ubbf8\uc9c0\ub97c \uace0\ud574\uc0c1\ub3c4\ub85c \ubcc0\ud658\ud558\ub294 \ub370\uc5d0 GAN\uc744 \ud65c\uc6a9\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/li>\n<\/ul>\n
3.1 \ucd5c\uadfc\uc758 GAN \uc5f0\uad6c \ub3d9\ud5a5<\/h3>\n
\ucd5c\uadfc \uc5f0\uad6c\uc5d0\uc11c\ub294 GAN\uc758 \ud559\uc2b5\uc744 \uc548\uc815\ud654\ud558\uae30 \uc704\ud55c \ub2e4\uc591\ud55c \uc811\uadfc \ubc29\uc2dd\uc774 \uc81c\uc548\ub418\uace0 \uc788\uc2b5\ub2c8\ub2e4. \uc608\ub97c \ub4e4\uc5b4, Wasserstein GAN(WGAN)\uc740 \uc190\uc2e4 \ud568\uc218\uc758 \uc548\uc815\uc131\uc744 \uac1c\uc120\ud558\uc5ec \ubaa8\ub4dc \ubd95\uad34\ub97c \ubc29\uc9c0\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.<\/p>\n
4. \uacb0\ub860<\/h2>\n
GAN\uc740 \uc774\ubbf8\uc9c0 \uc0dd\uc131 \ubc0f \ubcc0\ud658\uc5d0\uc11c \uc911\uc694\ud55c \uc5ed\ud560\uc744 \ud558\ub294 \ubaa8\ub378\ub85c, \ud30c\uc774\ud1a0\uce58\uc640 \uac19\uc740 \ud504\ub808\uc784\uc6cc\ud06c\ub97c \ud1b5\ud574 \uc27d\uac8c \uad6c\ud604\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. GAN\uc740 \uc55e\uc73c\ub85c\ub3c4 \ub2e4\uc591\ud55c \ubd84\uc57c\uc5d0\uc11c \uacc4\uc18d \ubc1c\uc804\ud560 \uac83\uc774\uba70, \ub525\ub7ec\ub2dd\uc758 \uacbd\uacc4\ub97c \ub113\ud788\ub294 \ub370 \uae30\uc5ec\ud560 \uac83\uc785\ub2c8\ub2e4.<\/p>\n
<\/body><\/p>\n","protected":false},"excerpt":{"rendered":"