- 🍨 本文为🔗365天深度学习训练营 中的学习记录博客
- 🍖 原作者：K同学啊 | 接辅导、项目定制
- 🚀 文章来源：K同学的学习圈子
本人电脑配置
Python 3.8.0
Pytorch 1.8.1
torchvision 1.8.1+ cuda10.2
前期准备
1. 设置GPU/CPU
本次是在gpu上对网络进行训练和测试，先识别设备，判断设备类型。
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(device)
2. 导入数据
本次采用的数据集为K同学提供的咖啡豆数据集。下载数据到主目录的文件夹的7-data文件夹里，下面是数据集的一些可视化结果。
data_dir = './7-data/'
data_dir = pathlib.Path(data_dir)
data_paths = list(data_dir.glob("*"))
classNames = [str(path).split("\\")[1] for path in data_paths]
print(classNames)
data_transform = transforms.Compose([
transforms.Resize([224, 224]),
transforms.ToTensor(),
transforms.Normalize(
mean=[0.485,0.456,0.406],
std = [0.229, 0.224, 0.225])
])
total_data = datasets.ImageFolder("./7-data/",transform=data_transform)
print(total_data.class_to_idx)
train_size = int(0.8*len(total_data))
test_size = len(total_data)-train_size
train_dataset, test_dataset = torch.utils.data.random_split(total_data, [train_size, test_size])
batch_size = 32
train_dl = torch.utils.data.DataLoader(train_dataset, batch_size=batch_size, shuffle = True)
test_dl = torch.utils.data.DataLoader(test_dataset, batch_size=batch_size, shuffle = True)
构建网络模型
1.自行搭建模型
自行搭建VGG16模型。
import torch.nn.functional as F
class vgg16(nn.Module):
def __init__(self):
super(vgg16, self).__init__()
# conv block 1
self.block1 = nn.Sequential(
nn.Conv2d(3, 64, kernel_size=(3,3), stride=(1, 1), padding=(1,1)),
nn.ReLU(),
nn.Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
nn.ReLU(),
nn.MaxPool2d(kernel_size=(2,2), stride=(2, 2))
)
self.block2 = nn.Sequential(
nn.Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1,1)),
nn.ReLU(),
nn.Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding= (1, 1)),
nn.ReLU(),
nn.MaxPool2d(kernel_size=(2,2), stride=(2,2))
)
self.block3 = nn.Sequential(
nn.Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
nn.ReLU(),
nn.Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
nn.ReLU(),
nn.Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
nn.ReLU(),
nn.MaxPool2d(kernel_size=(2, 2), stride=(2, 2))
)
self.block4 = nn.Sequential(
nn.Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
nn.ReLU(),
nn.Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
nn.ReLU(),
nn.Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1,1)),
nn.ReLU(),
nn.MaxPool2d(kernel_size=(2, 2), stride=(2, 2))
)
self.block5 = nn.Sequential(
nn.Conv2d(512, 512, kernel_size=(3,3), stride=(1,1), padding=(1, 1)),
nn.ReLU(),
nn.Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
nn.ReLU(),
nn.Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1)),
nn.ReLU(),
nn.MaxPool2d(kernel_size=(2, 2), stride=(2, 2))
)
self.classifier = nn.Sequential(
nn.Linear(in_features=512*7*7, out_features=4096),
nn.ReLU(),
nn.Dropout(p=0.5),
nn.Linear(in_features=4096, out_features=4096),
nn.ReLU(),
nn.Dropout(p=0.5),
nn.Linear(in_features=4096, out_features=4)
)
def forward(self, x):
x = self.block1(x)
x = self.block2(x)
x = self.block3(x)
x = self.block4(x)
x = self.block5(x)
x = torch.flatten(x, start_dim=1)
return x
device = "cuda" if torch.cuda.is_available() else "cpu"
model = vgg16().to(device)
也可以选择调用官方提供的vgg16和预训练模型，仅训练分类器，参照week6中改进模型，得到的模型结构如下
from torchvision import models
class VGGnet(nn.Module):
def __init__(self,feature_extract=True,num_classes=5):
super(VGGnet, self).__init__()
model = models.vgg16(pretrained=True)
self.features = model.features
set_parameter_requires_grad(self.features, feature_extract)#固定特征提取层参数
self.avgpool=model.avgpool
self.classifier = nn.Sequential(
nn.Linear(512 * 7 * 7, 512), #512 * 7 * 7不能改变 ，由VGG16网络决定的，第二个参数为神经元个数可以微调
nn.ReLU(True),
nn.Dropout(),
nn.Linear(512, 128),
nn.ReLU(True),
nn.Dropout(),
nn.Linear(128, len(classNames)),
)
def forward(self, x):
x = self.features(x)
x = self.avgpool(x)
x = x.view(x.size(0), 512*7*7)
out=self.classifier(x)
return out
def set_parameter_requires_grad(model, feature_extracting):
if feature_extracting:
for param in model.parameters():
param.requires_grad = False
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
model=VGGnet().to(device)
统计模型参数量并展示。
import torchsummary as summary
summary.summary(model, (3, 224, 224))
3. 编写训练函数
设置损失函数，这里采用的交叉熵损失函数。
# 训练循环
learn_rate = 1e-4 # 初始学习率
lambda1 = lambda epoch: 0.92 ** (epoch // 4)
loss_fn = nn.CrossEntropyLoss()
optimizer = torch.optim.SGD(model.parameters(), lr=learn_rate)
scheduler = torch.optim.lr_scheduler.LambdaLR(optimizer, lr_lambda=lambda1) #选定调整方法
# train model
def train(dataloader, model, loss_fn, optimizer):
size = len(dataloader.dataset)
num_batches = len(dataloader)
train_loss, train_acc = 0, 0
for X, y in dataloader:
X, y = X.to(device), y.to(device)
pred = model(X)
loss = loss_fn(pred, y)
optimizer.zero_grad()
loss.backward()
optimizer.step()
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
4. 编写测试函数
当不进行训练时，停止梯度更新，节省计算内存消耗。
# Test
def test(dataloader, model, loss_fn):
size = len(dataloader.dataset)
num_batches = len(dataloader)
test_loss, test_acc = 0, 0
with torch.no_grad():
for imgs, target in dataloader:
imgs, target = imgs.to(device), target.to(device)
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
5. 主函数
设置迭代epoch次数，这里设定为40，并记录训练误差、精度，测试误差、精度。在这里采用了学习率自适应的方式，即学习率随着epoch的增加而逐渐减小，。
import copy
optimizer = torch.optim.Adam(model.parameters(), lr= 1e-4)
loss_fn = nn.CrossEntropyLoss() # 创建损失函数
epochs = 40
train_loss = []
train_acc = []
test_loss = []
test_acc = []
best_acc = 0 # 设置一个最佳准确率，作为最佳模型的判别指标
for epoch in range(epochs):
model.train()
epoch_train_acc, epoch_train_loss = train(train_dl, model, loss_fn, optimizer)
model.eval()
epoch_test_acc, epoch_test_loss = test(test_dl, model, loss_fn)
# 保存最佳模型到 best_model
if epoch_test_acc > best_acc:
best_acc = epoch_test_acc
best_model = copy.deepcopy(model)
train_acc.append(epoch_train_acc)
train_loss.append(epoch_train_loss)
test_acc.append(epoch_test_acc)
test_loss.append(epoch_test_loss)
# 获取当前的学习率
lr = optimizer.state_dict()['param_groups'][0]['lr']
template = ('Epoch:{:2d}, Train_acc:{:.1f}%, Train_loss:{:.3f}, Test_acc:{:.1f}%, Test_loss:{:.3f}, Lr:{:.2E}')
print(template.format(epoch+1, epoch_train_acc*100, epoch_train_loss,
epoch_test_acc*100, epoch_test_loss, lr))
# 保存最佳模型到文件中
PATH = './best_model.pth' # 保存的参数文件名
torch.save(model.state_dict(), PATH)
结果总结
（1）采用自己构建的VGG16网络进行训练得到的训练结果如下。最终准确率可以达到96.7%，最好模型准确率为98.3%。
（2）调用官方的VGG16并再次训练分类器，得到的准确率为99.6%。模型的参数、训练结果和每个epoch的结果如下。
----------------------------------------------------------------
Layer (type) Output Shape Param #
================================================================
Conv2d-1 [-1, 64, 224, 224] 1,792
ReLU-2 [-1, 64, 224, 224] 0
Conv2d-3 [-1, 64, 224, 224] 36,928
ReLU-4 [-1, 64, 224, 224] 0
MaxPool2d-5 [-1, 64, 112, 112] 0
Conv2d-6 [-1, 128, 112, 112] 73,856
ReLU-7 [-1, 128, 112, 112] 0
Conv2d-8 [-1, 128, 112, 112] 147,584
ReLU-9 [-1, 128, 112, 112] 0
MaxPool2d-10 [-1, 128, 56, 56] 0
Conv2d-11 [-1, 256, 56, 56] 295,168
ReLU-12 [-1, 256, 56, 56] 0
Conv2d-13 [-1, 256, 56, 56] 590,080
ReLU-14 [-1, 256, 56, 56] 0
Conv2d-15 [-1, 256, 56, 56] 590,080
ReLU-16 [-1, 256, 56, 56] 0
MaxPool2d-17 [-1, 256, 28, 28] 0
Conv2d-18 [-1, 512, 28, 28] 1,180,160
ReLU-19 [-1, 512, 28, 28] 0
Conv2d-20 [-1, 512, 28, 28] 2,359,808
ReLU-21 [-1, 512, 28, 28] 0
Conv2d-22 [-1, 512, 28, 28] 2,359,808
ReLU-23 [-1, 512, 28, 28] 0
MaxPool2d-24 [-1, 512, 14, 14] 0
Conv2d-25 [-1, 512, 14, 14] 2,359,808
ReLU-26 [-1, 512, 14, 14] 0
Conv2d-27 [-1, 512, 14, 14] 2,359,808
ReLU-28 [-1, 512, 14, 14] 0
Conv2d-29 [-1, 512, 14, 14] 2,359,808
ReLU-30 [-1, 512, 14, 14] 0
MaxPool2d-31 [-1, 512, 7, 7] 0
AdaptiveAvgPool2d-32 [-1, 512, 7, 7] 0
Linear-33 [-1, 512] 12,845,568
ReLU-34 [-1, 512] 0
Dropout-35 [-1, 512] 0
Linear-36 [-1, 128] 65,664
ReLU-37 [-1, 128] 0
Dropout-38 [-1, 128] 0
Linear-39 [-1, 4] 516
================================================================
Total params: 27,626,436
Trainable params: 12,911,748
Non-trainable params: 14,714,688
----------------------------------------------------------------
Input size (MB): 0.57
Forward/backward pass size (MB): 218.60
Params size (MB): 105.39
Estimated Total Size (MB): 324.56
----------------------------------------------------------------
Epoch: 1, Train_acc:63.0%, Train_loss:0.987, Test_acc:95.4%, Test_loss:0.444, Lr:1.00E-04
Epoch: 2, Train_acc:89.6%, Train_loss:0.391, Test_acc:98.8%, Test_loss:0.166, Lr:1.00E-04
Epoch: 3, Train_acc:94.6%, Train_loss:0.207, Test_acc:99.2%, Test_loss:0.096, Lr:1.00E-04
Epoch: 4, Train_acc:97.2%, Train_loss:0.123, Test_acc:99.2%, Test_loss:0.065, Lr:1.00E-04
Epoch: 5, Train_acc:98.5%, Train_loss:0.083, Test_acc:99.6%, Test_loss:0.057, Lr:1.00E-04
Epoch: 6, Train_acc:99.1%, Train_loss:0.061, Test_acc:98.3%, Test_loss:0.056, Lr:1.00E-04
Epoch: 7, Train_acc:98.5%, Train_loss:0.054, Test_acc:98.8%, Test_loss:0.039, Lr:1.00E-04
Epoch: 8, Train_acc:99.5%, Train_loss:0.036, Test_acc:98.8%, Test_loss:0.039, Lr:1.00E-04
Epoch: 9, Train_acc:99.3%, Train_loss:0.031, Test_acc:98.8%, Test_loss:0.037, Lr:1.00E-04
Epoch:10, Train_acc:99.7%, Train_loss:0.028, Test_acc:97.5%, Test_loss:0.040, Lr:1.00E-04
Epoch:11, Train_acc:99.5%, Train_loss:0.023, Test_acc:97.9%, Test_loss:0.036, Lr:1.00E-04
Epoch:12, Train_acc:99.9%, Train_loss:0.014, Test_acc:98.8%, Test_loss:0.030, Lr:1.00E-04
Epoch:13, Train_acc:100.0%, Train_loss:0.011, Test_acc:99.6%, Test_loss:0.026, Lr:1.00E-04
Epoch:14, Train_acc:99.8%, Train_loss:0.015, Test_acc:99.2%, Test_loss:0.025, Lr:1.00E-04
Epoch:15, Train_acc:99.9%, Train_loss:0.012, Test_acc:99.6%, Test_loss:0.025, Lr:1.00E-04
Epoch:16, Train_acc:99.8%, Train_loss:0.010, Test_acc:99.6%, Test_loss:0.024, Lr:1.00E-04
Epoch:17, Train_acc:100.0%, Train_loss:0.007, Test_acc:100.0%, Test_loss:0.019, Lr:1.00E-04
Epoch:18, Train_acc:99.9%, Train_loss:0.007, Test_acc:98.8%, Test_loss:0.024, Lr:1.00E-04
Epoch:19, Train_acc:100.0%, Train_loss:0.005, Test_acc:99.6%, Test_loss:0.022, Lr:1.00E-04
Epoch:20, Train_acc:100.0%, Train_loss:0.005, Test_acc:99.6%, Test_loss:0.024, Lr:1.00E-04
Epoch:21, Train_acc:99.8%, Train_loss:0.007, Test_acc:98.8%, Test_loss:0.025, Lr:1.00E-04
Epoch:22, Train_acc:99.9%, Train_loss:0.006, Test_acc:99.6%, Test_loss:0.018, Lr:1.00E-04
Epoch:23, Train_acc:99.9%, Train_loss:0.007, Test_acc:98.3%, Test_loss:0.039, Lr:1.00E-04
Epoch:24, Train_acc:99.9%, Train_loss:0.006, Test_acc:98.8%, Test_loss:0.025, Lr:1.00E-04
Epoch:25, Train_acc:100.0%, Train_loss:0.003, Test_acc:98.8%, Test_loss:0.020, Lr:1.00E-04
Epoch:26, Train_acc:100.0%, Train_loss:0.003, Test_acc:98.3%, Test_loss:0.021, Lr:1.00E-04
Epoch:27, Train_acc:100.0%, Train_loss:0.002, Test_acc:98.8%, Test_loss:0.023, Lr:1.00E-04
Epoch:28, Train_acc:100.0%, Train_loss:0.001, Test_acc:99.2%, Test_loss:0.020, Lr:1.00E-04
Epoch:29, Train_acc:100.0%, Train_loss:0.003, Test_acc:98.8%, Test_loss:0.023, Lr:1.00E-04
Epoch:30, Train_acc:100.0%, Train_loss:0.002, Test_acc:98.8%, Test_loss:0.022, Lr:1.00E-04
Epoch:31, Train_acc:100.0%, Train_loss:0.002, Test_acc:99.2%, Test_loss:0.017, Lr:1.00E-04
Epoch:32, Train_acc:100.0%, Train_loss:0.002, Test_acc:99.2%, Test_loss:0.019, Lr:1.00E-04
Epoch:33, Train_acc:99.9%, Train_loss:0.002, Test_acc:98.8%, Test_loss:0.031, Lr:1.00E-04
Epoch:34, Train_acc:100.0%, Train_loss:0.002, Test_acc:98.8%, Test_loss:0.026, Lr:1.00E-04
Epoch:35, Train_acc:100.0%, Train_loss:0.002, Test_acc:98.8%, Test_loss:0.023, Lr:1.00E-04
Epoch:36, Train_acc:99.9%, Train_loss:0.003, Test_acc:99.2%, Test_loss:0.024, Lr:1.00E-04
Epoch:37, Train_acc:100.0%, Train_loss:0.002, Test_acc:98.8%, Test_loss:0.036, Lr:1.00E-04
Epoch:38, Train_acc:100.0%, Train_loss:0.002, Test_acc:98.8%, Test_loss:0.024, Lr:1.00E-04
Epoch:39, Train_acc:99.8%, Train_loss:0.004, Test_acc:98.8%, Test_loss:0.048, Lr:1.00E-04
Epoch:40, Train_acc:100.0%, Train_loss:0.003, Test_acc:99.6%, Test_loss:0.016, Lr:1.00E-04
Done
预测结果是：Dark
1.0 0.020511331531452015
扫一扫
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
