一、前期准备
1. 设置GPU
import torch from torch import nn import torchvision from torchvision import transforms,datasets,models import matplotlib.pyplot as plt import os,PIL,pathlib
device = torch.device("cuda" if torch.cuda.is_available() else "cpu") device
device(type='cuda')
2. 导入数据
data_dir = './data/' data_dir = pathlib.Path(data_dir) image_count = len(list(data_dir.glob('*/*/*.jpg'))) print("图片总数为：",image_count)
图片总数为： 578
classNames = [str(path).split('\\')[2] for path in data_dir.glob('train/*/')] classNames
['adidas', 'nike']
roses= list(data_dir.glob('train/nike/*.jpg')) PIL.Image.open(str(roses[0]))
3. 数据增强 解决过拟合
train_transforms = transforms.Compose([ transforms.Resize([224, 224]), transforms.RandomRotation(45),#随机旋转，-45到45度之间随机选 # transforms.CenterCrop(224),#从中心开始裁剪 transforms.RandomHorizontalFlip(p=0.5),#随机水平翻转 选择一个概率概率 # transforms.RandomVerticalFlip(p=0.5),#随机垂直翻转 # transforms.ColorJitter(brightness=0.2, contrast=0.1, saturation=0.1, hue=0.1),#参数1为亮度，参数2为对比度，参数3为饱和度，参数4为色相 # transforms.RandomGrayscale(p=0.025),#概率转换成灰度率，3通道就是R=G=B transforms.ToTensor(), transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])#均值，标准差 ]) test_transforms = transforms.Compose([ transforms.Resize([224, 224]), transforms.ToTensor(), transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]) ])
batch_size = 32 train_dataset = datasets.ImageFolder('./data/train/', transform = train_transforms) test_dataset = datasets.ImageFolder('./data/test/', transform = test_transforms) train_dl = torch.utils.data.DataLoader(train_dataset, batch_size=batch_size, shuffle=True, num_workers=1) test_dl = torch.utils.data.DataLoader(test_dataset, batch_size=batch_size, shuffle=True, num_workers=1)
classNames = train_dataset.classes train_dataset.class_to_idx
{'adidas': 0, 'nike': 1}
4. 数据可视化
imgs, labels = next(iter(train_dl)) imgs.shape
import numpy as np # 指定图片大小，图像大小为20宽、5高的绘图(单位为英寸inch) plt.figure(figsize=(20, 5)) for i, imgs in enumerate(imgs[:20]): npimg = imgs.numpy().transpose((1,2,0)) npimg = npimg * np.array((0.229, 0.224, 0.225)) + np.array((0.485, 0.456, 0.406)) npimg = npimg.clip(0, 1) # 将整个figure分成2行10列，绘制第i+1个子图。 plt.subplot(2, 10, i+1) plt.imshow(npimg) plt.axis('off')
for X,y in test_dl: print('Shape of X [N, C, H, W]:', X.shape) print('Shape of y:', y.shape) break
Shape of X [N, C, H, W]: torch.Size([32, 3, 224, 224])
Shape of y: torch.Size([32])
二、构建CNN网络
2.1 搭建简单网络
搭建简单网络后发现由于数据量少导致过拟合，数据增强后最高准确率84%，说明模型不够好，选择改用Resnet18+迁移学习：
2.2 迁移学习
2.2.1 调用resnet18和预训练模型、冻结参数
feature_extract = True # 冻结参数 def set_parameter_requires_grad(model, feature_extracting): if feature_extracting: for param in model.parameters(): param.requires_grad = False # 修改输出层 def initialize_model(num_classes, feature_extract, use_pretrained=True): model_ft = models.resnet18(pretrained=use_pretrained) set_parameter_requires_grad(model_ft, feature_extract) num_ftrs = model_ft.fc.in_features model_ft.fc = nn.Linear(num_ftrs, num_classes) input_size = 32 return model_ft, input_size
model_ft, input_size = initialize_model(2, feature_extract, use_pretrained=True) model_ft = model_ft.to(device) model_ft
略
2.2.2 取出输出层参数
取出输出层参数 后面用于训练更新
# 设置训练哪些层 params_to_update = model_ft.parameters() print("Params to learn:") if feature_extract: # 自己只训练输出层 params_to_update = [] for name,param in model_ft.named_parameters(): if param.requires_grad == True: params_to_update.append(param) print("\t",name) else: for name,param in model_ft.named_parameters(): if param.requires_grad == True: print("\t",name)
Params to learn:
fc.weight
fc.bias
三、训练模型
3.1 设置超参数
动态学习率
# 优化器设置 optimizer = torch.optim.Adam(params_to_update, lr=1e-4)#要训练啥参数，你来定 scheduler = torch.optim.lr_scheduler.StepLR(optimizer, step_size=5, gamma=0.92)#学习率每7个epoch衰减成原来的1/10 loss_fn = nn.CrossEntropyLoss()
# def adjust_learning_rate(optimizer, epoch, start_lr): # # 每2个 epoch衰减到原来的0.98 # lr = start_lr * (0.92 ** (epoch //2)) # for param_group in optimizer.param_groups: # param_group['lr'] = lr # optimizer = torch.optim.Adam(params_to_update,lr=1e-4)
3.2 编写训练函数
# 训练循环 def train(dataloader, model, loss_fn, optimizer): size = len(dataloader.dataset) # 训练集的大小，一共900张图片 num_batches = len(dataloader) # 批次数目，29（900/32） train_loss, train_acc = 0, 0 # 初始化训练损失和正确率 for X, y in dataloader: # 获取图片及其标签 X, y = X.to(device), y.to(device) # 计算预测误差 pred = model(X) # 网络输出 loss = loss_fn(pred, y) # 计算网络输出和真实值之间的差距，targets为真实值，计算二者差值即为损失 # 反向传播 optimizer.zero_grad() # grad属性归零 loss.backward() # 反向传播 optimizer.step() # 每一步自动更新 # 记录acc与loss train_acc += (pred.argmax(1) == y).type(torch.float).sum().item() train_loss += loss.item() train_acc /= size train_loss /= num_batches return train_acc, train_loss
3.3 编写测试函数
def test (dataloader, model, loss_fn): size = len(dataloader.dataset) # 测试集的大小，一共10000张图片 num_batches = len(dataloader) # 批次数目，8（255/32=8，向上取整） test_loss, test_acc = 0, 0 # 当不进行训练时，停止梯度更新，节省计算内存消耗 with torch.no_grad(): for imgs, target in dataloader: imgs, target = imgs.to(device), target.to(device) # 计算loss target_pred = model(imgs) loss = loss_fn(target_pred, target) test_loss += loss.item() test_acc += (target_pred.argmax(1) == target).type(torch.float).sum().item() test_acc /= size test_loss /= num_batches return test_acc, test_loss
3.4 正式训练
3.4.1 训练输出层
epochs = 20 train_loss = [] train_acc = [] test_loss = [] test_acc = [] best_acc = 0 filename='checkpoint.pth' for epoch in range(epochs): model_ft.train() epoch_train_acc, epoch_train_loss = train(train_dl, model_ft, loss_fn, optimizer) scheduler.step()#学习率衰减 model_ft.eval() epoch_test_acc, epoch_test_loss = test(test_dl, model_ft, loss_fn) # 保存最优模型 if epoch_test_acc > best_acc: best_acc = epoch_test_acc state = { 'state_dict': model_ft.state_dict(),#字典里key就是各层的名字，值就是训练好的权重 'best_acc': best_acc, 'optimizer' : optimizer.state_dict(), } torch.save(state, filename) train_acc.append(epoch_train_acc) train_loss.append(epoch_train_loss) test_acc.append(epoch_test_acc) test_loss.append(epoch_test_loss) template = ('Epoch:{:2d}, Train_acc:{:.1f}%, Train_loss:{:.3f}, Test_acc:{:.1f}%，Test_loss:{:.3f}') print(template.format(epoch+1, epoch_train_acc*100, epoch_train_loss, epoch_test_acc*100, epoch_test_loss)) print('Done') print('best_acc：',best_acc)
Epoch:17, Train_acc:65.9%, Train_loss:0.630, Test_acc:67.1%，Test_loss:0.615
Epoch:18, Train_acc:66.1%, Train_loss:0.613, Test_acc:64.5%，Test_loss:0.599
Epoch:19, Train_acc:63.7%, Train_loss:0.636, Test_acc:65.8%，Test_loss:0.579
Epoch:20, Train_acc:66.3%, Train_loss:0.612, Test_acc:65.8%，Test_loss:0.583
Done
best_acc： 0.6593625498007968
3.4.2 训练所有层
for param in model_ft.parameters(): param.requires_grad = True # 再继续训练所有的参数，学习率调小一点 optimizer = torch.optim.Adam(model_ft.parameters(), lr=1e-4) scheduler = torch.optim.lr_scheduler.StepLR(optimizer, step_size=5, gamma=0.92) # 损失函数 criterion = nn.CrossEntropyLoss()
# 加载之前训练好的权重参数 checkpoint = torch.load(filename) best_acc = checkpoint['best_acc'] model_ft.load_state_dict(checkpoint['state_dict'])
epochs = 20 train_loss = [] train_acc = [] test_loss = [] test_acc = [] best_acc = 0 filename='best_resnet18.pth' for epoch in range(epochs): model_ft.train() epoch_train_acc, epoch_train_loss = train(train_dl, model_ft, loss_fn, optimizer) scheduler.step()#学习率衰减 model_ft.eval() epoch_test_acc, epoch_test_loss = test(test_dl, model_ft, loss_fn) # 保存最优模型 if epoch_test_acc > best_acc: best_acc = epoch_test_acc state = { 'state_dict': model_ft.state_dict(),#字典里key就是各层的名字，值就是训练好的权重 'best_acc': best_acc, 'optimizer' : optimizer.state_dict(), } torch.save(state, filename) train_acc.append(epoch_train_acc) train_loss.append(epoch_train_loss) test_acc.append(epoch_test_acc) test_loss.append(epoch_test_loss) template = ('Epoch:{:2d}, Train_acc:{:.1f}%, Train_loss:{:.3f}, Test_acc:{:.1f}%，Test_loss:{:.3f}') print(template.format(epoch+1, epoch_train_acc*100, epoch_train_loss, epoch_test_acc*100, epoch_test_loss)) print('Done') print('best_acc：',best_acc)
Epoch:18, Train_acc:99.8%, Train_loss:0.010, Test_acc:86.8%，Test_loss:0.398
Epoch:19, Train_acc:99.0%, Train_loss:0.031, Test_acc:93.4%，Test_loss:0.203
Epoch:20, Train_acc:99.2%, Train_loss:0.019, Test_acc:93.4%，Test_loss:0.184
Done
best_acc： 0.9342105263157895
四、结果可视化
加载训练好的模型
model_ft, input_size = initialize_model(2, feature_extract, use_pretrained=True) # GPU模式 model_ft = model_ft.to(device) # 保存文件的名字 filename='best_resnet18.pth' # 加载模型 checkpoint = torch.load(filename) best_acc = checkpoint['best_acc'] model_ft.load_state_dict(checkpoint['state_dict'])
结果可视化
import matplotlib.pyplot as plt #隐藏警告 import warnings warnings.filterwarnings("ignore") #忽略警告信息 plt.rcParams['font.sans-serif'] = ['SimHei'] # 用来正常显示中文标签 plt.rcParams['axes.unicode_minus'] = False # 用来正常显示负号 plt.rcParams['figure.dpi'] = 100 #分辨率 epochs_range = range(epochs) plt.figure(figsize=(12, 3)) plt.subplot(1, 2, 1) plt.plot(epochs_range, train_acc, label='Training Accuracy') plt.plot(epochs_range, test_acc, label='Test Accuracy') plt.legend(loc='lower right') plt.title('Training and Validation Accuracy') plt.subplot(1, 2, 2) plt.plot(epochs_range, train_loss, label='Training Loss') plt.plot(epochs_range, test_loss, label='Test Loss') plt.legend(loc='upper right') plt.title('Training and Validation Loss') plt.show()
测试模型
train_on_gpu = True # 得到一个batch的测试数据 imgs, labels = next(iter(train_dl)) # 进行预测 model_ft.eval() if train_on_gpu: output = model_ft(imgs.cuda()) else: output = model_ft(imgs) # 获得预测结果(概率最大的) _, preds_tensor = torch.max(output, 1) preds = np.squeeze(preds_tensor.numpy()) if not train_on_gpu else np.squeeze(preds_tensor.cpu().numpy()) preds
array([0, 1, 1, 0, 0, 1, 1, 1, 0, 0, 0, 1, 0, 0, 0, 0, 1, 0, 1, 1, 1, 1,
0, 0, 0, 0, 1, 0, 1, 0, 0, 1], dtype=int64)
import numpy as np # 指定图片大小，图像大小为20宽、5高的绘图(单位为英寸inch) plt.figure(figsize=(20, 10)) for idx, imgs in enumerate(imgs[:10]): #ax = fig.add_subplot(rows, columns, idx+1, xticks=[], yticks=[]) npimg = imgs.numpy().transpose((1,2,0)) npimg = npimg * np.array((0.229, 0.224, 0.225)) + np.array((0.485, 0.456, 0.406)) npimg = npimg.clip(0, 1) # 将整个figure分成2行10列，绘制第i+1个子图。 ax = plt.subplot(2, 5, idx+1) ax.set_title("{} ({})".format(classNames[preds[idx]], classNames[labels[idx]]), color=("green" if classNames[preds[idx]]==classNames[labels[idx]] else "red")) plt.imshow(npimg) plt.axis('off')
