\n \uc804\uc774 \ud559\uc2b5(Transfer Learning)\uc740 \uba38\uc2e0\ub7ec\ub2dd \ubc0f \ub525\ub7ec\ub2dd \ubd84\uc57c\uc5d0\uc11c \ub9e4\uc6b0 \uc911\uc694\ud55c \uae30\uc220\uc785\ub2c8\ub2e4. \uc774 \uae30\uc220\uc740 \ud55c \uc791\uc5c5\uc5d0 \ub300\ud574 \ud559\uc2b5\ud55c \uac00\uc911\uce58\ub098 \ud30c\ub77c\ubbf8\ud130\ub97c \ub2e4\ub978 \uc720\uc0ac\ud55c \uc791\uc5c5\uc5d0 \uc7ac\uc0ac\uc6a9\ud558\ub294 \uacfc\uc815\uc744 \ub9d0\ud569\ub2c8\ub2e4. \uc804\uc774 \ud559\uc2b5\uc744 \ud1b5\ud574 \uc0d8\ud50c\uc758 \uc218\uac00 \uc801\uac70\ub098, \uc0c8\ub85c\uc6b4 \ub370\uc774\ud130\uc14b\uc744 \uc0ac\uc6a9\ud560 \ub54c \ub9ce\uc740 \uc2dc\uac04\uacfc \uc790\uc6d0\uc744 \uc808\uc57d\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.\n <\/p>\n
\n \ub370\uc774\ud130 \uc218\uc9d1 \ubc0f \ubaa8\ub378 \ud559\uc2b5\uc5d0\ub294 \ub9ce\uc740 \uc2dc\uac04\uacfc \ube44\uc6a9\uc774 \ud544\uc694\ud569\ub2c8\ub2e4. \ub530\ub77c\uc11c \uc774\ubbf8 \uc874\uc7ac\ud558\ub294 \ubaa8\ub378\uc5d0\uc11c \ud559\uc2b5\ud55c \uc9c0\uc2dd\uc744 \uc0c8\ub85c\uc6b4 \uc791\uc5c5\uc5d0 \ud65c\uc6a9\ud568\uc73c\ub85c\uc368 \ud6a8\uc728\uc131\uc744 \ub192\uc77c \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc608\ub97c \ub4e4\uc5b4, \uc774\ubbf8\uc9c0 \ubd84\ub958\ub97c \uc704\ud55c \ubaa8\ub378\uc774 \uc774\ubbf8 \ud559\uc2b5\ub41c \uacbd\uc6b0, \uc774\ub7ec\ud55c \ubaa8\ub378\uc744 \uc2dd\ubb3c \ubd84\ub958\uc640 \uac19\uc740 \uc720\uc0ac\ud55c \uc791\uc5c5\uc5d0 \ud65c\uc6a9\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4.\n <\/p>\n
\n \uc77c\ubc18\uc801\uc73c\ub85c \uc804\uc774 \ud559\uc2b5\uc740 \ub2e4\uc74c\uacfc \uac19\uc740 \ub2e8\uacc4\ub97c \ud3ec\ud568\ud569\ub2c8\ub2e4:\n <\/p>\n
\n \ud30c\uc774\ud1a0\uce58\ub294 \uc804\uc774 \ud559\uc2b5\uc744 \uc9c0\uc6d0\ud558\ub294 \ub2e4\uc591\ud55c \uae30\ub2a5\uc744 \uc81c\uacf5\ud569\ub2c8\ub2e4. \uc774\ub97c \ud1b5\ud574 \ubcf5\uc7a1\ud55c \ubaa8\ub378\uc744 \uc27d\uac8c \uc0ac\uc6a9\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc544\ub798\uc758 \uc608\uc81c\uc5d0\uc11c\ub294 \ud30c\uc774\ud1a0\uce58\uc758 torchvision \ub77c\uc774\ube0c\ub7ec\ub9ac\ub97c \uc774\uc6a9\ud574 \uc0ac\uc804 \ud6c8\ub828\ub41c \ubaa8\ub378\ub85c \uc774\ubbf8\uc9c0 \ubd84\ub958 \uc791\uc5c5\uc744 \uc218\ud589\ud558\ub294 \uacfc\uc815\uc744 \uc124\uba85\ud558\uaca0\uc2b5\ub2c8\ub2e4.\n <\/p>\n
\n \uc774\ubbf8\uc9c0 \ub370\uc774\ud130\uc14b\uc744 \ub85c\ub529\ud558\uace0 \uc804\ucc98\ub9ac\ud558\ub294 \ubc29\ubc95\uc744 \uc124\uba85\ud569\ub2c8\ub2e4. \uc5ec\uae30\uc11c\ub294 CIFAR-10 \ub370\uc774\ud130\uc14b\uc744 \uc0ac\uc6a9\ud558\uaca0\uc2b5\ub2c8\ub2e4.\n <\/p>\n
\nimport torch\nimport torchvision\nimport torchvision.transforms as transforms\n\n# \ub370\uc774\ud130 \uc804\ucc98\ub9ac\ntransform = transforms.Compose([\n transforms.Resize((224, 224)),\n transforms.ToTensor(),\n])\n\n# CIFAR-10 \ub370\uc774\ud130\uc14b \ub85c\ub4dc\ntrainset = torchvision.datasets.CIFAR10(root='.\/data', train=True,\n download=True, transform=transform)\ntrainloader = torch.utils.data.DataLoader(trainset, batch_size=64,\n shuffle=True, num_workers=2)\n\ntestset = torchvision.datasets.CIFAR10(root='.\/data', train=False,\n download=True, transform=transform)\ntestloader = torch.utils.data.DataLoader(testset, batch_size=64,\n shuffle=False, num_workers=2)\n <\/code><\/pre>\n4.2 \uc0ac\uc804 \ud6c8\ub828\ub41c \ubaa8\ub378 \ub85c\ub4dc<\/h3>\n
\n \ud30c\uc774\ud1a0\uce58\uc758 torchvision \ubaa8\ub378\uc5d0\uc11c \uc0ac\uc804 \ud6c8\ub828\ub41c ResNet18 \ubaa8\ub378\uc744 \ubd88\ub7ec\uc624\ub294 \ubc29\ubc95\uc744 \uc124\uba85\ud569\ub2c8\ub2e4.\n <\/p>\n
\nimport torchvision.models as models\n\n# \uc0ac\uc804 \ud6c8\ub828\ub41c \ubaa8\ub378 \ub85c\ub4dc\nmodel = models.resnet18(pretrained=True)\n\n# \ub9c8\uc9c0\ub9c9 \ub808\uc774\uc5b4 \uc218\uc815\nnum_classes = 10 # CIFAR-10\uc758 \ud074\ub798\uc2a4 \uc218\nmodel.fc = torch.nn.Linear(model.fc.in_features, num_classes)\n <\/code><\/pre>\n4.3 \uc190\uc2e4 \ud568\uc218 \ubc0f \ucd5c\uc801\ud654\uae30 \uc815\uc758<\/h3>\n
\n \ub2e4\uc911 \ud074\ub798\uc2a4 \ubd84\ub958 \ubb38\uc81c\ub97c \uc704\ud574 \uc190\uc2e4 \ud568\uc218\uc640 \ucd5c\uc801\ud654 \uc54c\uace0\ub9ac\uc998\uc744 \uc815\uc758\ud569\ub2c8\ub2e4.\n <\/p>\n
\nimport torch.optim as optim\n\ncriterion = torch.nn.CrossEntropyLoss() # \uc190\uc2e4 \ud568\uc218\noptimizer = optim.SGD(model.parameters(), lr=0.001, momentum=0.9) # \ucd5c\uc801\ud654 \uc54c\uace0\ub9ac\uc998\n <\/code><\/pre>\n4.4 \ubaa8\ub378 \ud6c8\ub828<\/h3>\n
\n \ubaa8\ub378\uc744 \ud6c8\ub828\uc2dc\ud0a4\uae30 \uc704\ud55c \uc804\uccb4\uc801\uc778 \ucf54\ub4dc\uc640 \ubc29\ubc95\uc744 \uc124\uba85\ud569\ub2c8\ub2e4.\n <\/p>\n
\n# \ubaa8\ub378 \ud6c8\ub828\ndevice = torch.device(\"cuda:0\" if torch.cuda.is_available() else \"cpu\")\nmodel.to(device)\n\nfor epoch in range(10): # \uc5d0\ud3ec\ud06c \uc218 \uc870\uc815 \uac00\ub2a5\n running_loss = 0.0\n for i, data in enumerate(trainloader, 0):\n inputs, labels = data\n inputs, labels = inputs.to(device), labels.to(device)\n\n # \uacbd\ub7c9\ud654 \uacbd\uacfc\n optimizer.zero_grad()\n outputs = model(inputs)\n loss = criterion(outputs, labels)\n loss.backward()\n optimizer.step()\n\n running_loss += loss.item()\n if i % 100 == 99: # \ub9e4 100 \ubbf8\ub2c8 \ubc30\uce58\ub9c8\ub2e4 \ucd9c\ub825\n print(f'[Epoch {epoch + 1}, Batch {i + 1}] loss: {running_loss \/ 100:.3f}')\n running_loss = 0.0\n\nprint('Finished Training')\n <\/code><\/pre>\n4.5 \ubaa8\ub378 \ud3c9\uac00<\/h3>\n
\n \ud6c8\ub828\uc774 \uc644\ub8cc\ub41c \ubaa8\ub378\uc744 \ud3c9\uac00\ud558\ub294 \ubc29\ubc95\uc5d0 \ub300\ud574 \uc124\uba85\ud569\ub2c8\ub2e4. \ud14c\uc2a4\ud2b8 \ub370\uc774\ud130\uc14b\uc744 \uc0ac\uc6a9\ud558\uc5ec \ubaa8\ub378\uc758 \uc815\ud655\ub3c4\ub97c \uce21\uc815\ud569\ub2c8\ub2e4.\n <\/p>\n
\n# \ubaa8\ub378 \ud3c9\uac00\ncorrect = 0\ntotal = 0\nwith torch.no_grad():\n for data in testloader:\n images, labels = data\n images, labels = images.to(device), labels.to(device)\n outputs = model(images)\n _, predicted = torch.max(outputs.data, 1)\n total += labels.size(0)\n correct += (predicted == labels).sum().item()\n\nprint(f'Accuracy of the network on the 10000 test images: {100 * correct \/ total:.2f}%')\n <\/code><\/pre>\n5. \uacb0\ub860<\/h2>\n
\n \ubcf8 \uac15\uc88c\uc5d0\uc11c\ub294 \ub525\ub7ec\ub2dd\uc5d0\uc11c \uc804\uc774 \ud559\uc2b5\uc758 \uac1c\ub150\uacfc \ud30c\uc774\ud1a0\uce58\ub97c \uc774\uc6a9\ud55c \uad6c\ud604 \ubc29\ubc95\uc744 \uc0b4\ud3b4\ubcf4\uc558\uc2b5\ub2c8\ub2e4. \uc804\uc774 \ud559\uc2b5\uc740 \ub370\uc774\ud130\uac00 \ubd80\uc871\ud55c \uc0c1\ud669\uc5d0\uc11c\ub3c4 \uac15\ub825\ud55c \uc131\ub2a5\uc744 \ub0bc \uc218 \uc788\ub3c4\ub85d \ub3d5\ub294 \uc911\uc694\ud55c \uae30\uc220\uc785\ub2c8\ub2e4. \ub2e4\uc591\ud55c \uc0ac\uc804 \ud6c8\ub828\ub41c \ubaa8\ub378\uc744 \ud65c\uc6a9\ud558\uc5ec \uc6b0\ub9ac\ub294 \ubcf4\ub2e4 \uc27d\uac8c \uace0\uc131\ub2a5 \ubaa8\ub378\uc744 \uac1c\ubc1c\ud560 \uc218 \uc788\uc2b5\ub2c8\ub2e4. \uc55e\uc73c\ub85c\ub3c4 \uc774\ub7ec\ud55c \uc804\uc774 \ud559\uc2b5\uc744 \ud1b5\ud574 \ub354 \ub9ce\uc740 \ub525\ub7ec\ub2dd \uc5b4\ud50c\ub9ac\ucf00\uc774\uc158\uc774 \uac1c\ubc1c\ub418\uae30\ub97c \uae30\ub300\ud569\ub2c8\ub2e4.\n <\/p>\n
6. \ucc38\uace0 \uc790\ub8cc<\/h2>\n