pytorch-torchvision.ipynb

{"cells":[{"cell_type":"markdown","id":"fifteen-currency","metadata":{},"source":["## 1、关于torchvision\n","\n","`torchvision` 是`PyTorch`中专门用来处理图像的库。这个包中有四个大类。\n","\n","- torchvision.datasets\n","- torchvision.models\n","- torchvision.transforms"]},{"cell_type":"markdown","id":"bulgarian-advance","metadata":{},"source":["## 2、torchvision.datasets"]},{"cell_type":"markdown","id":"known-watson","metadata":{},"source":["内置数据集：https://pytorch.org/vision/0.8/datasets.html\n","\n","案例："]},{"cell_type":"code","execution_count":null,"id":"absolute-latter","metadata":{},"outputs":[],"source":["import torchvision\n","\n","dataset = torchvision.datasets.MNIST(root='./mnist/', train=True,transform=torchvision.transforms.ToTensor(),download=True)"]},{"cell_type":"markdown","id":"specialized-pepper","metadata":{"execution":{"iopub.execute_input":"2021-02-15T12:22:53.366550Z","iopub.status.busy":"2021-02-15T12:22:53.365970Z","iopub.status.idle":"2021-02-15T12:22:54.111059Z","shell.execute_reply":"2021-02-15T12:22:54.109368Z","shell.execute_reply.started":"2021-02-15T12:22:53.366486Z"}},"source":["## 3、torchvision.models"]},{"cell_type":"code","execution_count":null,"id":"backed-amino","metadata":{},"outputs":[],"source":["!mkdir resnet50\n","!wget https://www.flyai.com/m/resnet50_ibn_a-d9d0bb7b.pth"]},{"cell_type":"markdown","id":"dominant-college","metadata":{},"source":["内置模型：https://pytorch.org/vision/0.8/models.html\n","\n","模型加速下载链接：https://www.flyai.com/models\n","\n","案例：\n","\n","### 1-只用模型不用参数"]},{"cell_type":"code","execution_count":null,"id":"adequate-consent","metadata":{},"outputs":[],"source":["import torchvision.models as models\n","resnet50 = models.resnet50(progress=True)"]},{"cell_type":"markdown","id":"intense-oregon","metadata":{},"source":["### 2-使用模型和参数"]},{"cell_type":"code","execution_count":null,"id":"geological-quest","metadata":{},"outputs":[],"source":["import torchvision.models as models\n","resnet50 = models.resnet50(pretrained=True,progress=True)"]},{"cell_type":"markdown","id":"listed-hazard","metadata":{},"source":["### 3-加载EfficientNet网络\n","\n","参考：https://blog.csdn.net/qq_38410428/article/details/95074279"]},{"cell_type":"code","execution_count":null,"id":"extra-consciousness","metadata":{},"outputs":[],"source":["!pip install efficientnet_pytorch\n","!wget https://www.flyai.com/m/adv-efficientnet-b3-cdd7c0f4.pth\n","    \n","from efficientnet_pytorch import EfficientNet\n","# 只加载模型不用参数\n","model = EfficientNet.from_name('efficientnet-b3')"]},{"cell_type":"code","execution_count":null,"id":"different-revelation","metadata":{},"outputs":[],"source":["# 加载网络和参数\n","model = EfficientNet.from_pretrained('efficientnet-b0')"]},{"cell_type":"code","execution_count":null,"id":"tracked-transport","metadata":{},"outputs":[],"source":["# 查看最后几行输出中的 out_features 为多少分类\n","print(model)"]},{"cell_type":"code","execution_count":null,"id":"breathing-albania","metadata":{},"outputs":[],"source":["import torch.nn as nn\n","\n","# 修改最后的分类\n","feature = model._fc.in_features\n","model._fc = nn.Linear(in_features=feature,out_features=45,bias=True)"]},{"cell_type":"code","execution_count":null,"id":"logical-marijuana","metadata":{},"outputs":[],"source":["# 再次检查最后几行输出中的 out_features 是否为设定类别数\n","print(model)"]},{"cell_type":"markdown","id":"sized-occupation","metadata":{},"source":["## 4、torchvision.transforms\n","\n","transforms方法：https://pytorch.org/vision/0.8/transforms.html\n","\n","案例："]},{"cell_type":"code","execution_count":null,"id":"renewable-kansas","metadata":{},"outputs":[],"source":["from torchvision import transforms\n","\n","transform_data = transforms.Compose([transforms.ToTensor(),\n","                              transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)),\n","                             ])\n","\n","dataset = torchvision.datasets.MNIST(root='./mnist/', train=True,transform=transform_data,download=True)"]},{"cell_type":"markdown","id":"noted-entrepreneur","metadata":{"execution":{"iopub.execute_input":"2021-02-15T16:25:04.950649Z","iopub.status.busy":"2021-02-15T16:25:04.950090Z","iopub.status.idle":"2021-02-15T16:25:04.959481Z","shell.execute_reply":"2021-02-15T16:25:04.958267Z","shell.execute_reply.started":"2021-02-15T16:25:04.950589Z"}},"source":["## 5、实战\n","\n","参考：\n","1. https://www.cnblogs.com/wj-1314/p/9842719.html\n","2. https://blog.csdn.net/qq_36560894/article/details/104506272\n","3. https://blog.csdn.net/qq_36560894/article/details/104923543"]},{"cell_type":"code","execution_count":23,"id":"transsexual-commodity","metadata":{"execution":{"iopub.execute_input":"2021-02-15T16:58:14.439107Z","iopub.status.busy":"2021-02-15T16:58:14.438541Z","iopub.status.idle":"2021-02-15T16:58:14.446201Z","shell.execute_reply":"2021-02-15T16:58:14.444782Z","shell.execute_reply.started":"2021-02-15T16:58:14.439051Z"}},"outputs":[],"source":["import torch\n","import torchvision\n","import torchvision.transforms as transforms\n","import torch.nn as nn"]},{"cell_type":"code","execution_count":50,"id":"close-shooting","metadata":{"execution":{"iopub.execute_input":"2021-02-15T17:55:37.782347Z","iopub.status.busy":"2021-02-15T17:55:37.781797Z","iopub.status.idle":"2021-02-15T17:55:37.847559Z","shell.execute_reply":"2021-02-15T17:55:37.846549Z","shell.execute_reply.started":"2021-02-15T17:55:37.782286Z"}},"outputs":[],"source":["train_transform = transforms.Compose( \n","    [transforms.ToTensor(), \n","     transforms.Lambda(lambda x: x.repeat(3,1,1)),\n","     transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))] \n",")\n","test_transform = transforms.Compose( \n","    [transforms.ToTensor(), \n","     transforms.Lambda(lambda x: x.repeat(3,1,1)),\n","     transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))] \n",")\n","\n","trainset = torchvision.datasets.MNIST(root='./data', train=True, download=True, transform=train_transform)\n","trainloader = torch.utils.data.DataLoader(trainset, batch_size=64, shuffle=True, num_workers=0)\n","\n","testset = torchvision.datasets.MNIST(root='./data', train=False, download=True, transform=test_transform)\n","testloader = torch.utils.data.DataLoader(testset, batch_size=64, shuffle=False, num_workers=0)"]},{"cell_type":"code","execution_count":28,"id":"focused-danish","metadata":{"execution":{"iopub.execute_input":"2021-02-15T17:00:23.596533Z","iopub.status.busy":"2021-02-15T17:00:23.595996Z","iopub.status.idle":"2021-02-15T17:00:23.975939Z","shell.execute_reply":"2021-02-15T17:00:23.975076Z","shell.execute_reply.started":"2021-02-15T17:00:23.596474Z"}},"outputs":[{"name":"stderr","output_type":"stream","text":["Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).\n"]},{"data":{"image/png":"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\n","text/plain":["<Figure size 432x288 with 1 Axes>"]},"metadata":{"needs_background":"light"},"output_type":"display_data"},{"name":"stdout","output_type":"stream","text":["tensor([4, 3, 7, 9, 5, 0, 7, 7, 1, 2, 6, 8, 1, 3, 9, 5, 7, 5, 1, 4, 6, 9, 5, 3,\n","        5, 9, 0, 3, 8, 7, 8, 3, 3, 6, 7, 6, 8, 4, 1, 4, 2, 9, 1, 3, 2, 3, 3, 9,\n","        0, 6, 3, 4, 8, 6, 9, 6, 4, 3, 3, 8, 1, 5, 1, 4])\n"]}],"source":["import matplotlib.pyplot as plt\n","import numpy as np\n","%matplotlib inline\n","\n","# 如果是RGB图像请注释掉 transforms.Lambda(lambda x: x.repeat(3,1,1)), 否则保留该语句\n","def imshow(img):\n","    npimg = img.numpy()\n","    plt.imshow(np.transpose(npimg, (1, 2, 0)))\n","    plt.show()\n","\n","dataiter = iter(trainloader)\n","images, labels = dataiter.next()\n","imshow(torchvision.utils.make_grid(images))\n","print(labels)"]},{"cell_type":"code","execution_count":33,"id":"floating-heath","metadata":{"execution":{"iopub.execute_input":"2021-02-15T17:09:11.064445Z","iopub.status.busy":"2021-02-15T17:09:11.063883Z","iopub.status.idle":"2021-02-15T17:14:51.145020Z","shell.execute_reply":"2021-02-15T17:14:51.143382Z","shell.execute_reply.started":"2021-02-15T17:09:11.064387Z"}},"outputs":[{"name":"stderr","output_type":"stream","text":["Downloading: \"https://download.pytorch.org/models/resnet18-5c106cde.pth\" to /output/.torch/hub/checkpoints/resnet18-5c106cde.pth\n"]},{"data":{"application/vnd.jupyter.widget-view+json":{"model_id":"af7a5ad6b2f848a3a0b100cc83dde99e","version_major":2,"version_minor":0},"text/plain":["  0%|          | 0.00/44.7M [00:00<?, ?B/s]"]},"metadata":{},"output_type":"display_data"}],"source":["import torchvision.models as models\n","model = models.resnet18(pretrained=True,progress=True)"]},{"cell_type":"code","execution_count":34,"id":"secret-joseph","metadata":{"execution":{"iopub.execute_input":"2021-02-15T17:15:09.539097Z","iopub.status.busy":"2021-02-15T17:15:09.538512Z","iopub.status.idle":"2021-02-15T17:15:09.549271Z","shell.execute_reply":"2021-02-15T17:15:09.547843Z","shell.execute_reply.started":"2021-02-15T17:15:09.539032Z"}},"outputs":[{"name":"stdout","output_type":"stream","text":["ResNet(\n","  (conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)\n","  (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","  (relu): ReLU(inplace=True)\n","  (maxpool): MaxPool2d(kernel_size=3, stride=2, padding=1, dilation=1, ceil_mode=False)\n","  (layer1): Sequential(\n","    (0): BasicBlock(\n","      (conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      (relu): ReLU(inplace=True)\n","      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","    )\n","    (1): BasicBlock(\n","      (conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      (relu): ReLU(inplace=True)\n","      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","    )\n","  )\n","  (layer2): Sequential(\n","    (0): BasicBlock(\n","      (conv1): Conv2d(64, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)\n","      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      (relu): ReLU(inplace=True)\n","      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      (downsample): Sequential(\n","        (0): Conv2d(64, 128, kernel_size=(1, 1), stride=(2, 2), bias=False)\n","        (1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      )\n","    )\n","    (1): BasicBlock(\n","      (conv1): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      (relu): ReLU(inplace=True)\n","      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","    )\n","  )\n","  (layer3): Sequential(\n","    (0): BasicBlock(\n","      (conv1): Conv2d(128, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)\n","      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      (relu): ReLU(inplace=True)\n","      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      (downsample): Sequential(\n","        (0): Conv2d(128, 256, kernel_size=(1, 1), stride=(2, 2), bias=False)\n","        (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      )\n","    )\n","    (1): BasicBlock(\n","      (conv1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      (relu): ReLU(inplace=True)\n","      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","    )\n","  )\n","  (layer4): Sequential(\n","    (0): BasicBlock(\n","      (conv1): Conv2d(256, 512, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)\n","      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      (relu): ReLU(inplace=True)\n","      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      (downsample): Sequential(\n","        (0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)\n","        (1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      )\n","    )\n","    (1): BasicBlock(\n","      (conv1): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      (relu): ReLU(inplace=True)\n","      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","    )\n","  )\n","  (avgpool): AdaptiveAvgPool2d(output_size=(1, 1))\n","  (fc): Linear(in_features=512, out_features=1000, bias=True)\n",")\n"]}],"source":["print(model)"]},{"cell_type":"code","execution_count":36,"id":"divided-wildlife","metadata":{"execution":{"iopub.execute_input":"2021-02-15T17:17:23.834742Z","iopub.status.busy":"2021-02-15T17:17:23.834241Z","iopub.status.idle":"2021-02-15T17:17:23.841542Z","shell.execute_reply":"2021-02-15T17:17:23.840464Z","shell.execute_reply.started":"2021-02-15T17:17:23.834690Z"}},"outputs":[],"source":["# 修改最后的分类\n","fc_features = model.fc.in_features \n","model.fc = nn.Linear(fc_features, 10)"]},{"cell_type":"markdown","id":"peaceful-burst","metadata":{},"source":["参考：https://blog.csdn.net/weixin_36670529/article/details/105910572"]},{"cell_type":"code","execution_count":37,"id":"similar-problem","metadata":{"execution":{"iopub.execute_input":"2021-02-15T17:17:33.124033Z","iopub.status.busy":"2021-02-15T17:17:33.123477Z","iopub.status.idle":"2021-02-15T17:17:33.132897Z","shell.execute_reply":"2021-02-15T17:17:33.131462Z","shell.execute_reply.started":"2021-02-15T17:17:33.123976Z"}},"outputs":[{"name":"stdout","output_type":"stream","text":["ResNet(\n","  (conv1): Conv2d(3, 64, kernel_size=(7, 7), stride=(2, 2), padding=(3, 3), bias=False)\n","  (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","  (relu): ReLU(inplace=True)\n","  (maxpool): MaxPool2d(kernel_size=3, stride=2, padding=1, dilation=1, ceil_mode=False)\n","  (layer1): Sequential(\n","    (0): BasicBlock(\n","      (conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      (relu): ReLU(inplace=True)\n","      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","    )\n","    (1): BasicBlock(\n","      (conv1): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      (relu): ReLU(inplace=True)\n","      (conv2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn2): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","    )\n","  )\n","  (layer2): Sequential(\n","    (0): BasicBlock(\n","      (conv1): Conv2d(64, 128, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)\n","      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      (relu): ReLU(inplace=True)\n","      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      (downsample): Sequential(\n","        (0): Conv2d(64, 128, kernel_size=(1, 1), stride=(2, 2), bias=False)\n","        (1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      )\n","    )\n","    (1): BasicBlock(\n","      (conv1): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      (relu): ReLU(inplace=True)\n","      (conv2): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn2): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","    )\n","  )\n","  (layer3): Sequential(\n","    (0): BasicBlock(\n","      (conv1): Conv2d(128, 256, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)\n","      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      (relu): ReLU(inplace=True)\n","      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      (downsample): Sequential(\n","        (0): Conv2d(128, 256, kernel_size=(1, 1), stride=(2, 2), bias=False)\n","        (1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      )\n","    )\n","    (1): BasicBlock(\n","      (conv1): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn1): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      (relu): ReLU(inplace=True)\n","      (conv2): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn2): BatchNorm2d(256, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","    )\n","  )\n","  (layer4): Sequential(\n","    (0): BasicBlock(\n","      (conv1): Conv2d(256, 512, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)\n","      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      (relu): ReLU(inplace=True)\n","      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      (downsample): Sequential(\n","        (0): Conv2d(256, 512, kernel_size=(1, 1), stride=(2, 2), bias=False)\n","        (1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      )\n","    )\n","    (1): BasicBlock(\n","      (conv1): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn1): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","      (relu): ReLU(inplace=True)\n","      (conv2): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), bias=False)\n","      (bn2): BatchNorm2d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n","    )\n","  )\n","  (avgpool): AdaptiveAvgPool2d(output_size=(1, 1))\n","  (fc): Linear(in_features=512, out_features=10, bias=True)\n",")\n"]}],"source":["print(model)"]},{"cell_type":"code","execution_count":38,"id":"continent-therapist","metadata":{"execution":{"iopub.execute_input":"2021-02-15T17:24:19.557301Z","iopub.status.busy":"2021-02-15T17:24:19.556768Z","iopub.status.idle":"2021-02-15T17:24:19.567139Z","shell.execute_reply":"2021-02-15T17:24:19.565922Z","shell.execute_reply.started":"2021-02-15T17:24:19.557241Z"}},"outputs":[],"source":["criterion = torch.nn.CrossEntropyLoss()\n","optimizer = torch.optim.Adam(model.parameters(),lr=2e-4)"]},{"cell_type":"code","execution_count":47,"id":"disturbed-growth","metadata":{"execution":{"iopub.execute_input":"2021-02-15T17:43:48.239941Z","iopub.status.busy":"2021-02-15T17:43:48.239393Z","iopub.status.idle":"2021-02-15T17:54:06.137727Z","shell.execute_reply":"2021-02-15T17:54:06.136021Z","shell.execute_reply.started":"2021-02-15T17:43:48.239885Z"}},"outputs":[{"name":"stdout","output_type":"stream","text":["epoch 0,iter 20,train accuracy: 97.1094%   loss:  0.1031\n","epoch 0,iter 40,train accuracy: 96.9531%   loss:  0.0887\n","epoch 0,iter 60,train accuracy: 98.0469%   loss:  0.0593\n","epoch 0,iter 80,train accuracy: 98.5156%   loss:  0.0527\n","epoch 0,iter 100,train accuracy: 98.0469%   loss:  0.0599\n","epoch 0,iter 120,train accuracy: 97.7344%   loss:  0.0806\n","epoch 0,iter 140,train accuracy: 98.2812%   loss:  0.0551\n","epoch 0,iter 160,train accuracy: 98.4375%   loss:  0.0525\n","epoch 0,iter 180,train accuracy: 98.7500%   loss:  0.0434\n","epoch 0,iter 200,train accuracy: 97.8125%   loss:  0.0853\n","epoch 0,iter 220,train accuracy: 98.5156%   loss:  0.0625\n","epoch 0,iter 240,train accuracy: 98.4375%   loss:  0.0604\n","epoch 0,iter 260,train accuracy: 98.5156%   loss:  0.0565\n","epoch 0,iter 280,train accuracy: 97.7344%   loss:  0.0744\n","epoch 0,iter 300,train accuracy: 98.5156%   loss:  0.0483\n","epoch 0,iter 320,train accuracy: 98.3594%   loss:  0.0494\n","epoch 0,iter 340,train accuracy: 97.5781%   loss:  0.0934\n","epoch 0,iter 360,train accuracy: 98.8281%   loss:  0.0539\n","epoch 0,iter 380,train accuracy: 98.9062%   loss:  0.0373\n","epoch 0,iter 400,train accuracy: 97.8125%   loss:  0.0673\n","epoch 0,iter 420,train accuracy: 98.7500%   loss:  0.0513\n"]},{"ename":"KeyboardInterrupt","evalue":"","output_type":"error","traceback":["\u001b[0;31m---------------------------------------------------------------------------\u001b[0m","\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)","\u001b[0;32m<ipython-input-47-9e8d05a78ba9>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[1;32m     25\u001b[0m         \u001b[0mouts\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mmodel\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mimages\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     26\u001b[0m         \u001b[0mloss\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mcriterion\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mouts\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mlabels\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 27\u001b[0;31m         \u001b[0mloss\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mbackward\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m     28\u001b[0m         \u001b[0moptimizer\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mstep\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     29\u001b[0m         \u001b[0;31m# 输出状态\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n","\u001b[0;32m/usr/local/lib/python3.6/site-packages/torch/tensor.py\u001b[0m in \u001b[0;36mbackward\u001b[0;34m(self, gradient, retain_graph, create_graph)\u001b[0m\n\u001b[1;32m    219\u001b[0m                 \u001b[0mretain_graph\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mretain_graph\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    220\u001b[0m                 create_graph=create_graph)\n\u001b[0;32m--> 221\u001b[0;31m         \u001b[0mtorch\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mautograd\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mbackward\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mgradient\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mretain_graph\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mcreate_graph\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    222\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    223\u001b[0m     \u001b[0;32mdef\u001b[0m \u001b[0mregister_hook\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mhook\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n","\u001b[0;32m/usr/local/lib/python3.6/site-packages/torch/autograd/__init__.py\u001b[0m in \u001b[0;36mbackward\u001b[0;34m(tensors, grad_tensors, retain_graph, create_graph, grad_variables)\u001b[0m\n\u001b[1;32m    130\u001b[0m     Variable._execution_engine.run_backward(\n\u001b[1;32m    131\u001b[0m         \u001b[0mtensors\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mgrad_tensors_\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mretain_graph\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mcreate_graph\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 132\u001b[0;31m         allow_unreachable=True)  # allow_unreachable flag\n\u001b[0m\u001b[1;32m    133\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    134\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n","\u001b[0;31mKeyboardInterrupt\u001b[0m: "]}],"source":["import time\n","\n","# 保存每个epoch后的Accuracy Loss Val_Accuracy\n","# device = torch.device('cuda:0' if torch.cuda.is_available() else 'cpu')\n","device = torch.device('cpu')\n","Accuracy = []\n","Loss = []\n","Val_Accuracy = []\n","BEST_VAL_ACC = 0.\n","# 训练\n","since = time.time()\n","for epoch in range(10):\n","    train_loss = 0.\n","    train_accuracy = 0.\n","    run_accuracy = 0.\n","    run_loss =0.\n","    total = 0.\n","    model.train()\n","    for i,data in enumerate(trainloader,0):\n","        images, labels = data\n","        images = images.to(device)\n","        labels = labels.to(device)  \n","        # 经典四步\n","        optimizer.zero_grad()\n","        outs = model(images)\n","        loss = criterion(outs, labels)\n","        loss.backward()\n","        optimizer.step()\n","        # 输出状态\n","        total += labels.size(0)\n","        run_loss += loss.item()\n","        _,prediction = torch.max(outs,1)\n","        run_accuracy += (prediction == labels).sum().item()\n","        if i % 20 == 19:\n","            print('epoch {},iter {},train accuracy: {:.4f}%   loss:  {:.4f}'.format(epoch, i+1, 100*run_accuracy/(labels.size(0)*20), run_loss/20))\n","            train_accuracy += run_accuracy\n","            train_loss += run_loss\n","            run_accuracy, run_loss = 0., 0.\n","    Loss.append(train_loss/total)\n","    Accuracy.append(100*train_accuracy/total)\n","    # 可视化训练过程\n","    fig1, ax1 = plt.subplots(figsize=(11, 8))\n","    ax1.plot(range(0, epoch+1, 1), Accuracy)\n","    ax1.set_title(\"Average trainset accuracy vs epochs\")\n","    ax1.set_xlabel(\"Epoch\")\n","    ax1.set_ylabel(\"Avg. train. accuracy\")\n","    plt.savefig('Train_accuracy_vs_epochs.png')\n","    plt.clf()\n","    plt.close()\n","    \n","    fig2, ax2 = plt.subplots(figsize=(11, 8))\n","    ax2.plot(range(epoch+1), Loss)\n","    ax2.set_title(\"Average trainset loss vs epochs\")\n","    ax2.set_xlabel(\"Epoch\")\n","    ax2.set_ylabel(\"Current loss\")\n","    plt.savefig('loss_vs_epochs.png')\n","\n","    plt.clf()\n","    plt.close()\n","    # 验证\n","    acc = 0.\n","    model.eval()\n","    print('waitting for Val...')\n","    with torch.no_grad():\n","        accuracy = 0.\n","        total =0\n","        for data in valloader:\n","            images, labels = data\n","            images = images.to(device)\n","            labels = labels.to(device)\n","            out = model(images)\n","            _, prediction = torch.max(out, 1)\n","            total += labels.size(0)\n","            accuracy += (prediction == labels).sum().item()\n","            acc = 100.*accuracy/total\n","    print('epoch {}  The ValSet accuracy is {:.4f}% \\n'.format(epoch, acc))\n","    Val_Accuracy.append(acc)\n","    if acc > BEST_VAL_ACC:\n","        print('Find Better Model and Saving it...')\n","        if not os.path.isdir('checkpoint'):\n","            os.mkdir('checkpoint')\n","        torch.save(model.state_dict(), './checkpoint/VGG19_Cats_Dogs_hc.pth')\n","        BEST_VAL_ACC = acc\n","        print('Saved!')\n","    \n","    fig3, ax3 = plt.subplots(figsize=(11, 8))\n","\n","    ax3.plot(range(epoch+1),Val_Accuracy )\n","    ax3.set_title(\"Average Val accuracy vs epochs\")\n","    ax3.set_xlabel(\"Epoch\")\n","    ax3.set_ylabel(\"Current Val accuracy\")\n","\n","    plt.savefig('val_accuracy_vs_epoch.png')\n","    plt.close()\n","    time_elapsed = time.time() - since\n","    print('Training complete in {:.0f}m {:.0f}s'.format(time_elapsed // 60, time_elapsed%60))\n","    print('Now the best val Acc is {:.4f}%'.format(BEST_VAL_ACC))"]},{"cell_type":"code","execution_count":48,"id":"detected-rating","metadata":{"execution":{"iopub.execute_input":"2021-02-15T17:54:10.339078Z","iopub.status.busy":"2021-02-15T17:54:10.338495Z","iopub.status.idle":"2021-02-15T17:54:10.421082Z","shell.execute_reply":"2021-02-15T17:54:10.420191Z","shell.execute_reply.started":"2021-02-15T17:54:10.339013Z"}},"outputs":[],"source":["PATH = './resnet18.pth'\n","torch.save(model.state_dict(), PATH)"]},{"cell_type":"code","execution_count":54,"id":"yellow-forty","metadata":{"execution":{"iopub.execute_input":"2021-02-15T18:00:10.004094Z","iopub.status.busy":"2021-02-15T18:00:10.003596Z","iopub.status.idle":"2021-02-15T18:01:27.245736Z","shell.execute_reply":"2021-02-15T18:01:27.244436Z","shell.execute_reply.started":"2021-02-15T18:00:10.004042Z"}},"outputs":[{"name":"stdout","output_type":"stream","text":["Accuracy of the network on the 10000 test images: 98 %\n"]}],"source":["# 评估结果1\n","correct = 0.\n","total = 0.\n","with torch.no_grad():  # 不需要计算梯度\n","    for data in testloader:\n","        images, labels = data\n","        out = model(images)\n","        _,prediction = torch.max(out, 1)\n","        total += labels.size(0)\n","        correct += (prediction == labels).sum().item()\n","        \n","print('Accuracy of the network on the 10000 test images: %d %%' % (100 * correct/total))\n"]},{"cell_type":"code","execution_count":null,"id":"possible-shareware","metadata":{},"outputs":[],"source":[]}],"metadata":{"kernelspec":{"display_name":"Python 3","language":"python","name":"python3"},"language_info":{"codemirror_mode":{"name":"ipython","version":3},"file_extension":".py","mimetype":"text/x-python","name":"python","nbconvert_exporter":"python","pygments_lexer":"ipython3","version":"3.6.12"},"deepnote_notebook_id":"c832ce71-fe3f-4021-917d-a4c7c3e6bf49"},"nbformat":4,"nbformat_minor":5}