├── .gitignore ├── LICENSE ├── README.md ├── cnn.py ├── train.py └── train_iter.py /.gitignore: -------------------------------------------------------------------------------- 1 | # Byte-compiled / optimized / DLL files 2 | __pycache__/ 3 | *.py[cod] 4 | *$py.class 5 | 6 | # C extensions 7 | *.so 8 | 9 | # Distribution / packaging 10 | .Python 11 | build/ 12 | develop-eggs/ 13 | dist/ 14 | downloads/ 15 | eggs/ 16 | .eggs/ 17 | lib/ 18 | lib64/ 19 | parts/ 20 | sdist/ 21 | var/ 22 | wheels/ 23 | *.egg-info/ 24 | .installed.cfg 25 | *.egg 26 | MANIFEST 27 | 28 | # PyInstaller 29 | # Usually these files are written by a python script from a template 30 | # before PyInstaller builds the exe, so as to inject date/other infos into it. 31 | *.manifest 32 | *.spec 33 | 34 | # Installer logs 35 | pip-log.txt 36 | pip-delete-this-directory.txt 37 | 38 | # Unit test / coverage reports 39 | htmlcov/ 40 | .tox/ 41 | .coverage 42 | .coverage.* 43 | .cache 44 | nosetests.xml 45 | coverage.xml 46 | *.cover 47 | .hypothesis/ 48 | .pytest_cache/ 49 | 50 | # Translations 51 | *.mo 52 | *.pot 53 | 54 | # Django stuff: 55 | *.log 56 | local_settings.py 57 | db.sqlite3 58 | 59 | # Flask stuff: 60 | instance/ 61 | .webassets-cache 62 | 63 | # Scrapy stuff: 64 | .scrapy 65 | 66 | # Sphinx documentation 67 | docs/_build/ 68 | 69 | # PyBuilder 70 | target/ 71 | 72 | # Jupyter Notebook 73 | .ipynb_checkpoints 74 | 75 | # pyenv 76 | .python-version 77 | 78 | # celery beat schedule file 79 | celerybeat-schedule 80 | 81 | # SageMath parsed files 82 | *.sage.py 83 | 84 | # Environments 85 | .env 86 | .venv 87 | env/ 88 | venv/ 89 | ENV/ 90 | env.bak/ 91 | venv.bak/ 92 | 93 | # Spyder project settings 94 | .spyderproject 95 | .spyproject 96 | 97 | # Rope project settings 98 | .ropeproject 99 | 100 | # mkdocs documentation 101 | /site 102 | 103 | # mypy 104 | .mypy_cache/ 105 | -------------------------------------------------------------------------------- /LICENSE: -------------------------------------------------------------------------------- 1 | MIT License 2 | 3 | Copyright (c) 2019 liqy2019 4 | 5 | Permission is hereby granted, free of charge, to any person obtaining a copy 6 | of this software and associated documentation files (the "Software"), to deal 7 | in the Software without restriction, including without limitation the rights 8 | to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 9 | copies of the Software, and to permit persons to whom the Software is 10 | furnished to do so, subject to the following conditions: 11 | 12 | The above copyright notice and this permission notice shall be included in all 13 | copies or substantial portions of the Software. 14 | 15 | THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 16 | IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 17 | FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE 18 | AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 19 | LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 20 | OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 21 | SOFTWARE. 22 | -------------------------------------------------------------------------------- /README.md: -------------------------------------------------------------------------------- 1 | # CNN-for-MNIST 2 | 使用CNN实现对手写数字的识别 3 | # Requirement 4 | Using the PyTorch framework for neural networks 5 | *PyTorch version 1.2.0 6 | *Python version 3.7.3 7 | # Usage 8 | train.py为对数据集的单次EPOCH训练 9 | train_iter.py为对数据集的多次EPOCH训练 10 | 11 | 如果觉得不错！点个star！ 12 | -------------------------------------------------------------------------------- /cnn.py: -------------------------------------------------------------------------------- 1 | from torch import nn 2 | 3 | 4 | 5 | class CNN(nn.Module): 6 | 7 | def __init__(self): 8 | 9 | super(CNN, self).__init__() 10 | 11 | self.layer1 = nn.Sequential( 12 | 13 | nn.Conv2d(1, 25, kernel_size=3), 14 | 15 | nn.BatchNorm2d(25), 16 | 17 | nn.ReLU(inplace=True) 18 | 19 | ) 20 | 21 | 22 | 23 | self.layer2 = nn.Sequential( 24 | 25 | nn.MaxPool2d(kernel_size=2, stride=2) 26 | 27 | ) 28 | 29 | 30 | 31 | self.layer3 = nn.Sequential( 32 | 33 | nn.Conv2d(25, 50, kernel_size=3), 34 | 35 | nn.BatchNorm2d(50), 36 | 37 | nn.ReLU(inplace=True) 38 | 39 | ) 40 | 41 | 42 | 43 | self.layer4 = nn.Sequential( 44 | 45 | nn.MaxPool2d(kernel_size=2, stride=2) 46 | 47 | ) 48 | 49 | 50 | 51 | self.fc = nn.Sequential( 52 | 53 | nn.Linear(50 * 5 * 5, 1024), 54 | 55 | nn.ReLU(inplace=True), 56 | 57 | nn.Linear(1024, 128), 58 | 59 | nn.ReLU(inplace=True), 60 | 61 | nn.Linear(128, 10) 62 | 63 | ) 64 | 65 | 66 | 67 | 68 | def forward(self, x): 69 | 70 | x = self.layer1(x) 71 | 72 | x = self.layer2(x) 73 | 74 | x = self.layer3(x) 75 | 76 | x = self.layer4(x) 77 | 78 | x = x.view(x.size(0), -1) 79 | 80 | x = self.fc(x) 81 | 82 | return x 83 | -------------------------------------------------------------------------------- /train.py: -------------------------------------------------------------------------------- 1 | '''================================================= 2 | @IDE ：Pycharm 3 | @Author ：Qingyong Li 4 | @Date ：2019/11/22 5 | 6 | ==================================================''' 7 | import torch 8 | 9 | from torch import nn, optim 10 | 11 | from torch.autograd import Variable 12 | 13 | from torch.utils.data import DataLoader 14 | 15 | from torchvision import datasets, transforms 16 | 17 | import cnn 18 | 19 | 20 | # 定义一些超参数 21 | 22 | batch_size = 64 23 | 24 | learning_rate = 0.02 25 | 26 | num_epoches = 20 27 | 28 | 29 | 30 | # 数据预处理。transforms.ToTensor()将图片转换成PyTorch中处理的对象Tensor,并且进行标准化（数据在0~1之间） 31 | 32 | # transforms.Normalize()做归一化。它进行了减均值，再除以标准差。两个参数分别是均值和标准差 33 | 34 | # transforms.Compose()函数则是将各种预处理的操作组合到了一起 35 | 36 | data_tf = transforms.Compose( 37 | 38 | [transforms.ToTensor(), 39 | 40 | transforms.Normalize([0.5], [0.5]) 41 | ]) 42 | 43 | 44 | 45 | # 数据集的下载器 46 | 47 | train_dataset = datasets.MNIST( 48 | 49 | root='./data', train=True, transform=data_tf, download=True) 50 | 51 | test_dataset = datasets.MNIST(root='./data', train=False, transform=data_tf) 52 | 53 | train_loader = DataLoader(train_dataset, batch_size=batch_size, shuffle=True) 54 | 55 | test_loader = DataLoader(test_dataset, batch_size=batch_size, shuffle=False) 56 | 57 | 58 | 59 | # 选择模型 60 | 61 | model = cnn.CNN() 62 | 63 | # model = net.Activation_Net(28 * 28, 300, 100, 10) 64 | 65 | # model = net.Batch_Net(28 * 28, 300, 100, 10) 66 | 67 | if torch.cuda.is_available(): 68 | 69 | model = model.cuda() 70 | 71 | 72 | 73 | # 定义损失函数和优化器 74 | 75 | criterion = nn.CrossEntropyLoss() 76 | 77 | optimizer = optim.SGD(model.parameters(), lr=learning_rate) 78 | 79 | 80 | 81 | # 训练模型 82 | 83 | epoch = 0 84 | 85 | for data in train_loader: 86 | 87 | img, label = data 88 | 89 | if torch.cuda.is_available(): 90 | 91 | img = img.cuda() 92 | 93 | label = label.cuda() 94 | 95 | else: 96 | 97 | img = Variable(img) 98 | 99 | label = Variable(label) 100 | 101 | out = model(img) 102 | 103 | loss = criterion(out, label) 104 | 105 | print_loss = loss.data.item() 106 | 107 | optimizer.zero_grad() 108 | 109 | loss.backward() 110 | 111 | optimizer.step() 112 | 113 | epoch+=1 114 | 115 | if epoch%50 == 0: 116 | 117 | print('epoch: {}, loss: {:.4}'.format(epoch, loss.data.item())) 118 | 119 | 120 | # 保存和加载整个模型 121 | 122 | torch.save(model, 'CNN_for_MNIST.pth') 123 | 124 | 125 | 126 | 127 | # 模型评估 128 | 129 | model.eval() 130 | 131 | eval_loss = 0 132 | 133 | eval_acc = 0 134 | 135 | for data in test_loader: 136 | 137 | img, label = data 138 | img = Variable(img) 139 | 140 | if torch.cuda.is_available(): 141 | 142 | img = img.cuda() 143 | label = label.cuda() 144 | 145 | out = model(img) 146 | loss = criterion(out, label) 147 | eval_loss += loss.data.item()*label.size(0) 148 | _, pred = torch.max(out, 1) 149 | num_correct = (pred == label).sum() 150 | eval_acc += num_correct.item() 151 | 152 | print('Test Loss: {:.6f}, Acc: {:.6f}'.format( 153 | 154 | eval_loss / (len(test_dataset)), 155 | 156 | eval_acc / (len(test_dataset)) 157 | 158 | )) 159 | -------------------------------------------------------------------------------- /train_iter.py: -------------------------------------------------------------------------------- 1 | '''================================================= 2 | @IDE ：Pycharm 3 | @Author ：Qingyong Li 4 | @Date ：2019/11/22 5 | ==================================================''' 6 | import torch 7 | from torch import nn, optim 8 | from torch.autograd import Variable 9 | from torch.utils.data import DataLoader 10 | from torchvision import datasets, transforms 11 | 12 | import cnn 13 | 14 | # 定义一些超参数 15 | 16 | batch_size = 128 17 | learning_rate = 0.01 18 | num_epoches = 20 19 | 20 | # 数据预处理。transforms.ToTensor()将图片转换成PyTorch中处理的对象Tensor,并且进行标准化（数据在0~1之间） 21 | # transforms.Normalize()做归一化。它进行了减均值，再除以标准差。两个参数分别是均值和标准差 22 | # transforms.Compose()函数则是将各种预处理的操作组合到了一起 23 | 24 | data_tf = transforms.Compose( 25 | [transforms.ToTensor(), 26 | transforms.Normalize([0.5], [0.5])]) 27 | 28 | 29 | 30 | # 数据集的下载器 31 | 32 | train_dataset = datasets.MNIST(root='./data', train=True, transform=data_tf, download=True) 33 | 34 | test_dataset = datasets.MNIST(root='./data', train=False, transform=data_tf) 35 | 36 | train_loader = DataLoader(train_dataset, batch_size=batch_size, shuffle=True) 37 | 38 | test_loader = DataLoader(test_dataset, batch_size=batch_size, shuffle=False) 39 | 40 | 41 | 42 | # 选择模型 43 | 44 | model = cnn.CNN() 45 | 46 | # model = net.Activation_Net(28 * 28, 300, 100, 10) 47 | 48 | # model = net.Batch_Net(28 * 28, 300, 100, 10) 49 | 50 | if torch.cuda.is_available(): 51 | print("CUDA is available") 52 | model = model.cuda() 53 | 54 | 55 | 56 | # 定义损失函数和优化器 57 | 58 | criterion = nn.CrossEntropyLoss() 59 | 60 | optimizer = optim.SGD(model.parameters(), lr=learning_rate) 61 | 62 | 63 | 64 | # 训练模型 65 | for i in range(num_epoches): 66 | epoch = 0 67 | 68 | for data in train_loader: 69 | 70 | img, label = data 71 | 72 | # img = img.view(img.size(0), -1) 73 | 74 | img = Variable(img) 75 | 76 | if torch.cuda.is_available(): 77 | 78 | img = img.cuda() 79 | 80 | label = label.cuda() 81 | 82 | else: 83 | 84 | img = Variable(img) 85 | 86 | label = Variable(label) 87 | 88 | out = model(img) 89 | 90 | loss = criterion(out, label) 91 | 92 | print_loss = loss.data.item() 93 | optimizer.zero_grad() 94 | 95 | loss.backward() 96 | 97 | optimizer.step() 98 | 99 | epoch+=1 100 | 101 | #if epoch%50 == 0: 102 | 103 | #print('epoch: {}, loss: {:.4}'.format(epoch, loss.data.item())) 104 | 105 | 106 | 107 | # 模型评估 108 | 109 | model.eval() 110 | 111 | eval_loss = 0 112 | 113 | eval_acc = 0 114 | 115 | for data in test_loader: 116 | 117 | img, label = data 118 | 119 | # img = img.view(img.size(0), -1) 120 | 121 | img = Variable(img) 122 | 123 | if torch.cuda.is_available(): 124 | 125 | img = img.cuda() 126 | 127 | label = label.cuda() 128 | 129 | 130 | 131 | out = model(img) 132 | 133 | loss = criterion(out, label) 134 | 135 | eval_loss += loss.data.item()*label.size(0) 136 | 137 | _, pred = torch.max(out, 1) 138 | 139 | num_correct = (pred == label).sum() 140 | 141 | eval_acc += num_correct.item() 142 | print('EPOCH: ',i+1) 143 | print('Test Loss: {:.6f}, Acc: {:.6f}'.format( 144 | 145 | eval_loss / (len(test_dataset)), 146 | 147 | eval_acc / (len(test_dataset)) 148 | 149 | )) 150 | i+=1 151 | #保存模型 152 | torch.save(model, 'CNN_for_MNIST.pth') --------------------------------------------------------------------------------