网络概述：网络架构如下（编码中若忽略正则化准确率低）
文件夹内容
在datasets-car文件夹中包含三类文件夹，其中train和val文件夹内部格式一致，test文件夹直接为纯图组成：
程序1car_data_origin：
from torch.utils.data import DataLoader, Dataset
from torchvision import transforms as T
import os
from PIL import Image
class Car(Dataset):
def __init__(self, root, transforms=None):
imgs = []
labels = []
# 定义车辆类别和对应标签的字典
class_labels = {
"truck": 0,
"taxi": 1,
"minibus": 2,
"fire engine": 3,
"racing car": 4,
"SUV": 5,
"bus": 6,
"jeep": 7,
"family sedan": 8,
"heavy truck": 9
}
for class_name in os.listdir(root):
class_path = os.path.join(root, class_name)
if os.path.isdir(class_path):
label = class_labels.get(class_name, None)
if label is not None:
for img_name in os.listdir(class_path):
img_path = os.path.join(class_path, img_name)
if os.path.isfile(img_path):
imgs.append(img_path)
labels.append(label)
self.imgs = imgs
self.labels = labels
self.transforms = transforms
def __getitem__(self, index):
img_path = self.imgs[index]
label = self.labels[index]
data = Image.open(img_path)
if data.mode != "RGB":
data = data.convert("RGB")
if self.transforms is not None:
data = self.transforms(data)
return data, label
def __len__(self):
return len(self.imgs)
if __name__ == "__main__":
root = "./datasets/car/train"
train_dataset = Car(root)
train_dataloader = DataLoader(train_dataset, batch_size=32, shuffle=True)
for data, label in train_dataset:
print(data.size) # 直接访问元组的元素，不需要使用括号
程序二VGG16（每次卷积完后、激活函数前增加归一化操作）：
import torch
import torch.nn as nn
class VGG16(nn.Module):
def __init__(self):
super(VGG16, self).__init__()
self.conv_layers = nn.Sequential(
# 第一组卷积层和池化层
nn.Conv2d(3, 64, kernel_size=3, padding=1),
nn.BatchNorm2d(64), # 添加批量归一化
nn.ReLU(inplace=True),
nn.Conv2d(64, 64, kernel_size=3, padding=1),
nn.BatchNorm2d(64),
nn.ReLU(inplace=True),
nn.MaxPool2d(kernel_size=2, stride=2),
# 第二组卷积层和池化层
nn.Conv2d(64, 128, kernel_size=3, padding=1),
nn.BatchNorm2d(128),
nn.ReLU(inplace=True),
nn.Conv2d(128, 128, kernel_size=3, padding=1),
nn.BatchNorm2d(128),
nn.ReLU(inplace=True),
nn.MaxPool2d(kernel_size=2, stride=2),
# 第三组卷积层和池化层
nn.Conv2d(128, 256, kernel_size=3, padding=1),
nn.BatchNorm2d(256),
nn.ReLU(inplace=True),
nn.Conv2d(256, 256, kernel_size=3, padding=1),
nn.BatchNorm2d(256),
nn.ReLU(inplace=True),
nn.Conv2d(256, 256, kernel_size=3, padding=1),
nn.BatchNorm2d(256),
nn.ReLU(inplace=True),
nn.MaxPool2d(kernel_size=2, stride=2),
# 第四组卷积层和池化层
nn.Conv2d(256, 512, kernel_size=3, padding=1),
nn.BatchNorm2d(512),
nn.ReLU(inplace=True),
nn.Conv2d(512, 512, kernel_size=3, padding=1),
nn.BatchNorm2d(512),
nn.ReLU(inplace=True),
nn.Conv2d(512, 512, kernel_size=3, padding=1),
nn.BatchNorm2d(512),
nn.ReLU(inplace=True),
nn.MaxPool2d(kernel_size=2, stride=2),
# 第五组卷积层和池化层
nn.Conv2d(512, 512, kernel_size=3, padding=1),
nn.BatchNorm2d(512),
nn.ReLU(inplace=True),
nn.Conv2d(512, 512, kernel_size=3, padding=1),
nn.BatchNorm2d(512),
nn.ReLU(inplace=True),
nn.Conv2d(512, 512, kernel_size=3, padding=1),
nn.BatchNorm2d(512),
nn.ReLU(inplace=True),
nn.MaxPool2d(kernel_size=2, stride=2),
)
self.fc_layers = nn.Sequential(
# 全连接层
nn.Linear(512 * 7 * 7, 4096),
nn.BatchNorm1d(4096),
nn.ReLU(inplace=True),
nn.Dropout(),
nn.Linear(4096, 4096),
nn.BatchNorm1d(4096),
nn.ReLU(inplace=True),
nn.Dropout(),
nn.Linear(4096, 10) # 输出层，10个类别的分类
)
def forward(self, x):
x = self.conv_layers(x)
x = x.view(x.size(0), -1)
x = self.fc_layers(x)
return x
运行结果： 
程序三Train_VGG16：
import os
import time
import torch
import torch.nn as nn
import torch.optim as optim
from torch.utils.data import DataLoader
from torchvision import transforms
from VGG16 import VGG16
from car_data_origin import Car
import matplotlib.pyplot as plt
def main():
# 数据处理和加载
pipline_train = transforms.Compose([
transforms.RandomHorizontalFlip(),
transforms.Resize((224, 224)),
transforms.ToTensor(),
transforms.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225))
])
pipline_test = transforms.Compose([
transforms.Resize((224, 224)),
transforms.ToTensor(),
transforms.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225))
])
train_data = Car(root="E:/deeplearnig/VGG and Siamese(1)/datasets/car/train", transforms=pipline_train)
test_data = Car(root="E:/deeplearnig/VGG and Siamese(1)/datasets/car/val", transforms=pipline_test)
trainloader = DataLoader(train_data, batch_size=8, shuffle=True)
testloader = DataLoader(test_data, batch_size=4, shuffle=False)
# 创建模型，部署到GPU
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = VGG16().to(device)
# 损失函数和优化器
criterion = nn.CrossEntropyLoss()
optimizer = optim.Adam(model.parameters(), lr=0.001)
epochs = 100
Loss = []
Accuracy = []
for epoch in range(1, epochs + 1):
start_time = time.time()
train_loss, train_acc = train_runner(model, device, trainloader, optimizer, criterion, epoch)
test_acc, test_loss = test_runner(model, device, testloader, criterion)
end_time = time.time()
Loss.append(train_loss)
Accuracy.append(train_acc)
print(f"Epoch {epoch}/{epochs}, Time: {end_time - start_time:.2f}s")
print(f"Train Accuracy: {train_acc:.2f}%, Train Average Loss: {train_loss:.6f}")
print(f"Test Accuracy: {test_acc:.2f}%, Test Average Loss: {test_loss:.6f}\n")
print('Finished Training')
# 保存模型
savePath = "../models/"
if not os.path.exists(savePath):
os.mkdir(savePath)
torch.save(model.state_dict(), savePath + 'VGG16_car.pth')
print("Model saved.")
plt.subplot(2, 1, 1)
plt.plot(Loss)
plt.title('Loss')
plt.subplot(2, 1, 2)
plt.plot(Accuracy)
plt.title('Accuracy')
plt.show()
def train_runner(model, device, trainloader, optimizer, criterion, epoch):
# 训练模型
model.train()
total_loss = 0.0
correct = 0
total = 0
for batch_idx, (inputs, targets) in enumerate(trainloader):
inputs, targets = inputs.to(device), targets.to(device)
optimizer.zero_grad()
outputs = model(inputs)
loss = criterion(outputs, targets)
loss.backward()
optimizer.step()
total_loss += loss.item()
_, predicted = outputs.max(1)
total += targets.size(0)
correct += predicted.eq(targets).sum().item()
avg_loss = total_loss / len(trainloader)
accuracy = 100. * correct / total
return avg_loss, accuracy
def test_runner(model, device, testloader, criterion):
# 模型验证
model.eval()
correct = 0
total = 0
test_loss = 0.0
with torch.no_grad():
for data, target in testloader:
data, target = data.to(device), target.to(device)
output = model(data)
test_loss += criterion(output, target).item()
_, predicted = output.max(1)
total += target.size(0)
correct += predicted.eq(target).sum().item()
avg_loss = test_loss / len(testloader)
accuracy = 100. * correct / total
return accuracy, avg_loss
if __name__ == '__main__':
main()
运行结果：
C:\Users\Administrator\.conda\envs\pytorch\python.exe "E:\deeplearnig\VGG and Siamese(1)\Train_VGG16.py"
Epoch 1/100, Time: 44.18s
Train Accuracy: 12.64%, Train Average Loss: 2.950671
Test Accuracy: 17.00%, Test Average Loss: 2.664477
Epoch 2/100, Time: 42.79s
Train Accuracy: 22.93%, Train Average Loss: 2.730200
Test Accuracy: 32.00%, Test Average Loss: 2.620904
Epoch 3/100, Time: 43.07s
Train Accuracy: 28.14%, Train Average Loss: 2.692980
Test Accuracy: 27.00%, Test Average Loss: 2.449761
Epoch 4/100, Time: 42.63s
Train Accuracy: 33.86%, Train Average Loss: 2.554127
Test Accuracy: 30.50%, Test Average Loss: 2.129170
Epoch 5/100, Time: 42.90s
Train Accuracy: 36.57%, Train Average Loss: 2.447231
Test Accuracy: 40.00%, Test Average Loss: 1.961281
Epoch 6/100, Time: 43.05s
Train Accuracy: 37.07%, Train Average Loss: 2.512312
Test Accuracy: 50.50%, Test Average Loss: 1.510240
Epoch 7/100, Time: 42.84s
Train Accuracy: 42.50%, Train Average Loss: 2.291613
Test Accuracy: 59.00%, Test Average Loss: 1.390003
Epoch 8/100, Time: 42.55s
Train Accuracy: 41.57%, Train Average Loss: 2.409683
Test Accuracy: 41.00%, Test Average Loss: 3.413687
Epoch 9/100, Time: 42.38s
Train Accuracy: 47.07%, Train Average Loss: 2.114320
Test Accuracy: 57.00%, Test Average Loss: 2.988450
Epoch 10/100, Time: 42.89s
Train Accuracy: 48.43%, Train Average Loss: 2.164377
Test Accuracy: 54.00%, Test Average Loss: 3.032696
Epoch 11/100, Time: 43.06s
Train Accuracy: 48.50%, Train Average Loss: 2.084977
Test Accuracy: 46.00%, Test Average Loss: 2.505732
Epoch 12/100, Time: 43.27s
Train Accuracy: 48.93%, Train Average Loss: 2.008778
Test Accuracy: 56.00%, Test Average Loss: 1.591382
Epoch 13/100, Time: 42.66s
Train Accuracy: 53.07%, Train Average Loss: 1.823400
Test Accuracy: 59.00%, Test Average Loss: 1.575225
Epoch 14/100, Time: 42.58s
Train Accuracy: 56.07%, Train Average Loss: 1.714480
Test Accuracy: 59.50%, Test Average Loss: 1.320819
Epoch 15/100, Time: 43.01s
Train Accuracy: 56.00%, Train Average Loss: 1.713857
Test Accuracy: 59.00%, Test Average Loss: 1.701987
Epoch 16/100, Time: 42.74s
Train Accuracy: 51.93%, Train Average Loss: 2.148327
Test Accuracy: 43.50%, Test Average Loss: 2.414334
Epoch 17/100, Time: 43.49s
Train Accuracy: 53.00%, Train Average Loss: 1.708398
Test Accuracy: 60.50%, Test Average Loss: 1.448982
Epoch 18/100, Time: 43.62s
Train Accuracy: 56.29%, Train Average Loss: 1.650544
Test Accuracy: 51.50%, Test Average Loss: 2.163751
Epoch 19/100, Time: 43.37s
Train Accuracy: 60.36%, Train Average Loss: 1.448843
Test Accuracy: 68.00%, Test Average Loss: 1.161868
Epoch 20/100, Time: 43.07s
Train Accuracy: 63.00%, Train Average Loss: 1.407892
Test Accuracy: 60.00%, Test Average Loss: 1.858408
Epoch 21/100, Time: 47.64s
Train Accuracy: 66.50%, Train Average Loss: 1.220983
Test Accuracy: 55.00%, Test Average Loss: 4.102234
Epoch 22/100, Time: 43.61s
Train Accuracy: 66.36%, Train Average Loss: 1.267141
Test Accuracy: 71.00%, Test Average Loss: 1.201189
Epoch 23/100, Time: 43.05s
Train Accuracy: 69.36%, Train Average Loss: 1.198333
Test Accuracy: 62.50%, Test Average Loss: 1.538909
Epoch 24/100, Time: 42.85s
Train Accuracy: 70.79%, Train Average Loss: 1.110783
Test Accuracy: 69.00%, Test Average Loss: 1.423777
Epoch 25/100, Time: 43.27s
Train Accuracy: 69.50%, Train Average Loss: 1.070163
Test Accuracy: 62.50%, Test Average Loss: 1.775788
Epoch 26/100, Time: 45.39s
Train Accuracy: 70.79%, Train Average Loss: 1.001934
Test Accuracy: 66.00%, Test Average Loss: 1.426365
Epoch 27/100, Time: 42.90s
Train Accuracy: 72.50%, Train Average Loss: 0.941015
Test Accuracy: 63.50%, Test Average Loss: 1.966777
Epoch 28/100, Time: 42.77s
Train Accuracy: 74.57%, Train Average Loss: 0.919872
Test Accuracy: 67.50%, Test Average Loss: 1.406872
Epoch 29/100, Time: 42.82s
Train Accuracy: 76.71%, Train Average Loss: 0.759704
Test Accuracy: 67.00%, Test Average Loss: 1.410608
Epoch 30/100, Time: 42.87s
Train Accuracy: 80.71%, Train Average Loss: 0.665533
Test Accuracy: 67.50%, Test Average Loss: 1.323388
Epoch 31/100, Time: 43.18s
Train Accuracy: 80.14%, Train Average Loss: 0.633734
Test Accuracy: 67.50%, Test Average Loss: 1.191376
Epoch 32/100, Time: 42.54s
Train Accuracy: 80.21%, Train Average Loss: 0.703288
Test Accuracy: 66.50%, Test Average Loss: 1.574172
Epoch 33/100, Time: 43.16s
Train Accuracy: 80.71%, Train Average Loss: 0.648511
Test Accuracy: 68.50%, Test Average Loss: 1.253326
Epoch 34/100, Time: 42.75s
Train Accuracy: 82.86%, Train Average Loss: 0.586593
Test Accuracy: 66.00%, Test Average Loss: 1.300445
Epoch 35/100, Time: 43.32s
Train Accuracy: 83.29%, Train Average Loss: 0.534224
Test Accuracy: 68.50%, Test Average Loss: 1.600032
Epoch 36/100, Time: 43.07s
Train Accuracy: 85.50%, Train Average Loss: 0.525826
Test Accuracy: 68.50%, Test Average Loss: 1.450944
Epoch 37/100, Time: 42.20s
Train Accuracy: 85.43%, Train Average Loss: 0.475364
Test Accuracy: 72.00%, Test Average Loss: 1.250811
Epoch 38/100, Time: 42.45s
Train Accuracy: 84.43%, Train Average Loss: 0.585558
Test Accuracy: 67.50%, Test Average Loss: 1.663046
Epoch 39/100, Time: 42.30s
Train Accuracy: 85.79%, Train Average Loss: 0.485690
Test Accuracy: 68.50%, Test Average Loss: 1.362564
Epoch 40/100, Time: 42.59s
Train Accuracy: 85.57%, Train Average Loss: 0.457015
Test Accuracy: 71.00%, Test Average Loss: 1.482029
Epoch 41/100, Time: 42.82s
Train Accuracy: 85.79%, Train Average Loss: 0.522534
Test Accuracy: 72.00%, Test Average Loss: 1.352253
Epoch 42/100, Time: 42.80s
Train Accuracy: 87.79%, Train Average Loss: 0.398443
Test Accuracy: 72.00%, Test Average Loss: 1.256099
Epoch 43/100, Time: 42.44s
Train Accuracy: 88.36%, Train Average Loss: 0.429865
Test Accuracy: 74.50%, Test Average Loss: 1.474210
Epoch 44/100, Time: 42.87s
Train Accuracy: 88.14%, Train Average Loss: 0.416003
Test Accuracy: 73.50%, Test Average Loss: 1.274462
Epoch 45/100, Time: 43.02s
Train Accuracy: 87.64%, Train Average Loss: 0.452182
Test Accuracy: 74.00%, Test Average Loss: 1.450168
Epoch 46/100, Time: 43.64s
Train Accuracy: 89.57%, Train Average Loss: 0.378843
Test Accuracy: 72.00%, Test Average Loss: 1.351240
Epoch 47/100, Time: 43.05s
Train Accuracy: 91.00%, Train Average Loss: 0.289240
Test Accuracy: 72.50%, Test Average Loss: 1.386482
Epoch 48/100, Time: 42.51s
Train Accuracy: 90.71%, Train Average Loss: 0.307707
Test Accuracy: 78.00%, Test Average Loss: 1.279066
Epoch 49/100, Time: 42.66s
Train Accuracy: 91.57%, Train Average Loss: 0.299976
Test Accuracy: 74.00%, Test Average Loss: 1.304493
Epoch 50/100, Time: 42.62s
Train Accuracy: 88.86%, Train Average Loss: 0.429909
Test Accuracy: 73.00%, Test Average Loss: 1.382795
Epoch 51/100, Time: 42.46s
Train Accuracy: 91.21%, Train Average Loss: 0.306615
Test Accuracy: 76.00%, Test Average Loss: 1.309700
Epoch 52/100, Time: 42.42s
Train Accuracy: 91.00%, Train Average Loss: 0.286100
Test Accuracy: 72.00%, Test Average Loss: 1.635651
Epoch 53/100, Time: 42.21s
Train Accuracy: 90.64%, Train Average Loss: 0.357244
Test Accuracy: 72.50%, Test Average Loss: 1.313712
Epoch 54/100, Time: 42.36s
Train Accuracy: 93.57%, Train Average Loss: 0.220612
Test Accuracy: 76.00%, Test Average Loss: 1.292393
Epoch 55/100, Time: 42.95s
Train Accuracy: 91.50%, Train Average Loss: 0.284906
Test Accuracy: 73.00%, Test Average Loss: 1.575787
Epoch 56/100, Time: 42.63s
Train Accuracy: 92.21%, Train Average Loss: 0.275172
Test Accuracy: 74.00%, Test Average Loss: 1.624438
Epoch 57/100, Time: 43.36s
Train Accuracy: 91.93%, Train Average Loss: 0.294722
Test Accuracy: 71.00%, Test Average Loss: 1.263066
Epoch 58/100, Time: 42.36s
Train Accuracy: 92.86%, Train Average Loss: 0.254413
Test Accuracy: 77.00%, Test Average Loss: 1.346407
Epoch 59/100, Time: 42.28s
Train Accuracy: 93.29%, Train Average Loss: 0.226384
Test Accuracy: 74.00%, Test Average Loss: 1.384903
Epoch 60/100, Time: 44.00s
Train Accuracy: 91.21%, Train Average Loss: 0.352764
Test Accuracy: 76.00%, Test Average Loss: 1.272360
Epoch 61/100, Time: 54.58s
Train Accuracy: 91.79%, Train Average Loss: 0.306122
Test Accuracy: 76.00%, Test Average Loss: 1.465147
Epoch 62/100, Time: 43.22s
Train Accuracy: 93.00%, Train Average Loss: 0.261215
Test Accuracy: 75.00%, Test Average Loss: 1.587256
Epoch 63/100, Time: 42.37s
Train Accuracy: 90.36%, Train Average Loss: 0.348683
Test Accuracy: 71.00%, Test Average Loss: 1.548762
Epoch 64/100, Time: 42.73s
Train Accuracy: 92.29%, Train Average Loss: 0.318729
Test Accuracy: 76.00%, Test Average Loss: 1.787213
Epoch 65/100, Time: 42.29s
Train Accuracy: 94.21%, Train Average Loss: 0.205660
Test Accuracy: 76.50%, Test Average Loss: 1.267424
Epoch 66/100, Time: 42.07s
Train Accuracy: 95.29%, Train Average Loss: 0.181558
Test Accuracy: 71.50%, Test Average Loss: 1.549532
Epoch 67/100, Time: 42.14s
Train Accuracy: 95.07%, Train Average Loss: 0.167123
Test Accuracy: 72.50%, Test Average Loss: 1.525913
Epoch 68/100, Time: 41.84s
Train Accuracy: 94.43%, Train Average Loss: 0.211938
Test Accuracy: 75.50%, Test Average Loss: 1.643690
Epoch 69/100, Time: 41.91s
Train Accuracy: 93.86%, Train Average Loss: 0.204199
Test Accuracy: 76.50%, Test Average Loss: 1.277502
Epoch 70/100, Time: 42.11s
Train Accuracy: 94.57%, Train Average Loss: 0.192549
Test Accuracy: 72.00%, Test Average Loss: 1.818104
Epoch 71/100, Time: 42.13s
Train Accuracy: 93.29%, Train Average Loss: 0.305557
Test Accuracy: 74.00%, Test Average Loss: 1.566282
Epoch 72/100, Time: 42.02s
Train Accuracy: 93.29%, Train Average Loss: 0.245612
Test Accuracy: 72.50%, Test Average Loss: 1.571880
Epoch 73/100, Time: 42.07s
Train Accuracy: 94.07%, Train Average Loss: 0.190021
Test Accuracy: 77.00%, Test Average Loss: 1.300863
Epoch 74/100, Time: 42.09s
Train Accuracy: 93.86%, Train Average Loss: 0.220239
Test Accuracy: 79.00%, Test Average Loss: 1.434459
Epoch 75/100, Time: 42.46s
Train Accuracy: 94.79%, Train Average Loss: 0.200344
Test Accuracy: 76.00%, Test Average Loss: 1.651057
Epoch 76/100, Time: 42.53s
Train Accuracy: 94.50%, Train Average Loss: 0.219200
Test Accuracy: 74.50%, Test Average Loss: 1.448879
Epoch 77/100, Time: 42.81s
Train Accuracy: 95.21%, Train Average Loss: 0.186794
Test Accuracy: 77.50%, Test Average Loss: 1.706803
Epoch 78/100, Time: 42.01s
Train Accuracy: 94.93%, Train Average Loss: 0.167908
Test Accuracy: 77.00%, Test Average Loss: 1.343045
Epoch 79/100, Time: 42.11s
Train Accuracy: 95.00%, Train Average Loss: 0.157969
Test Accuracy: 78.00%, Test Average Loss: 1.484615
Epoch 80/100, Time: 42.20s
Train Accuracy: 93.36%, Train Average Loss: 0.269918
Test Accuracy: 75.50%, Test Average Loss: 1.431945
Epoch 81/100, Time: 42.15s
Train Accuracy: 94.64%, Train Average Loss: 0.220715
Test Accuracy: 74.50%, Test Average Loss: 1.453883
Epoch 82/100, Time: 42.17s
Train Accuracy: 95.07%, Train Average Loss: 0.174723
Test Accuracy: 78.50%, Test Average Loss: 1.574471
Epoch 83/100, Time: 42.12s
Train Accuracy: 95.36%, Train Average Loss: 0.193206
Test Accuracy: 75.50%, Test Average Loss: 1.527657
Epoch 84/100, Time: 43.10s
Train Accuracy: 94.43%, Train Average Loss: 0.236524
Test Accuracy: 69.50%, Test Average Loss: 1.732961
Epoch 85/100, Time: 52.68s
Train Accuracy: 95.43%, Train Average Loss: 0.147453
Test Accuracy: 72.00%, Test Average Loss: 1.825806
Epoch 86/100, Time: 42.43s
Train Accuracy: 94.43%, Train Average Loss: 0.247971
Test Accuracy: 79.00%, Test Average Loss: 1.411562
Epoch 87/100, Time: 42.41s
Train Accuracy: 95.64%, Train Average Loss: 0.149416
Test Accuracy: 80.00%, Test Average Loss: 1.355214
Epoch 88/100, Time: 42.32s
Train Accuracy: 96.50%, Train Average Loss: 0.154553
Test Accuracy: 78.00%, Test Average Loss: 1.720926
Epoch 89/100, Time: 42.99s
Train Accuracy: 96.64%, Train Average Loss: 0.144905
Test Accuracy: 77.00%, Test Average Loss: 1.885772
Epoch 90/100, Time: 42.34s
Train Accuracy: 95.71%, Train Average Loss: 0.149564
Test Accuracy: 74.50%, Test Average Loss: 1.985403
Epoch 91/100, Time: 41.87s
Train Accuracy: 95.71%, Train Average Loss: 0.143841
Test Accuracy: 76.00%, Test Average Loss: 1.520910
Epoch 92/100, Time: 42.20s
Train Accuracy: 95.21%, Train Average Loss: 0.193025
Test Accuracy: 74.50%, Test Average Loss: 1.806884
Epoch 93/100, Time: 42.28s
Train Accuracy: 95.29%, Train Average Loss: 0.177240
Test Accuracy: 76.50%, Test Average Loss: 1.725439
Epoch 94/100, Time: 42.14s
Train Accuracy: 96.64%, Train Average Loss: 0.163867
Test Accuracy: 70.00%, Test Average Loss: 2.068439
Epoch 95/100, Time: 42.56s
Train Accuracy: 97.36%, Train Average Loss: 0.108272
Test Accuracy: 74.00%, Test Average Loss: 1.719326
Epoch 96/100, Time: 41.91s
Train Accuracy: 96.21%, Train Average Loss: 0.134013
Test Accuracy: 74.50%, Test Average Loss: 1.552624
Epoch 97/100, Time: 43.43s
Train Accuracy: 97.29%, Train Average Loss: 0.076644
Test Accuracy: 76.50%, Test Average Loss: 1.303036
Epoch 98/100, Time: 43.42s
Train Accuracy: 97.14%, Train Average Loss: 0.099385
Test Accuracy: 75.50%, Test Average Loss: 1.585183
Epoch 99/100, Time: 42.50s
Train Accuracy: 95.29%, Train Average Loss: 0.149875
Test Accuracy: 74.50%, Test Average Loss: 1.940277
Epoch 100/100, Time: 42.87s
Train Accuracy: 95.00%, Train Average Loss: 0.201341
Test Accuracy: 73.50%, Test Average Loss: 1.784691
Finished Training
Model saved.程序四：（界面开发要求：先显示左上角为“图像分类”标题字样的对话框，界面上方为“原始图像”，界面左下角为“打开图像”按钮，右下角为“识别”按钮，点击“打开图像”会跳转到电脑文件夹去选择图片，选好的图片会呈现在界面中央，点击“识别”按钮，按钮右侧会显示红色的识别结果（该结果为键而不是数字代替）加感叹号）
import torch
import torch.nn.functional as F
import torchvision.transforms as transforms
from PIL import Image, ImageTk
import tkinter as tk
from tkinter import filedialog
from VGG16 import VGG16
def open_image():
fpath = filedialog.askopenfilename()
img = Image.open(fpath)
img = img.resize((300, 300)) # 调整图片大小以适应界面
img_tk = ImageTk.PhotoImage(img)
panel.configure(image=img_tk)
panel.image = img_tk # 保持图片对象的引用
def recognize():
img_tensor = transform(img)
img_tensor = img_tensor.to(device)
img_tensor = img_tensor.unsqueeze(0) # 图片扩展多一维，适配模型输入格式
# 预测
with torch.no_grad():
output = model(img_tensor)
probabilities = F.softmax(output, dim=1)
predicted_class = torch.argmax(probabilities, dim=1).item()
# 更新识别结果标签
label_result.config(text=f"识别结果：{class_labels[predicted_class]}!")
# 创建窗口
window = tk.Tk()
window.title("图像分类")
# 加载模型
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
model = VGG16()
model.load_state_dict(torch.load('../models/VGG16_car.pth'))
model = model.to(device)
model.eval() # 把模型转为test模式
# 创建标签
label_title = tk.Label(window, text="原始图像", font=("Arial", 16))
label_title.grid(row=0, column=0, columnspan=2, sticky="w", padx=10, pady=10)
# 创建图片显示面板
panel = tk.Label(window)
panel.grid(row=1, column=0, columnspan=2, padx=10, pady=10)
# 创建打开图像按钮
btn_open = tk.Button(window, text="打开图像", command=open_image)
btn_open.grid(row=2, column=0, padx=10, pady=10, sticky="w")
# 创建识别按钮
btn_recognize = tk.Button(window, text="识别", command=recognize)
btn_recognize.grid(row=2, column=1, padx=10, pady=10, sticky="e")
# 加载图片预处理变换
transform = transforms.Compose([
transforms.Resize((224, 224)), # 将图片尺寸resize到224x224
transforms.ToTensor(),
transforms.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225)) # 标准化
])
# 定义车辆类别和对应标签的字典
class_labels = {
0: "卡车",
1: "出租车",
2: "面包车",
3: "消防车",
4: "赛车",
5: "SUV",
6: "公共汽车",
7: "吉普车",
8: "家用轿车",
9: "重型卡车"
}
# 运行窗口
window.mainloop()
运行结果：
注意点：
如果程序二VGG16，不加归一层，其训练过程的准确率始终低于20%，且不稳定，不随训练次数增加呈上升态势，架构如下所示
import torch
import torch.nn as nn
class VGG16(nn.Module):
def __init__(self):
super(VGG16, self).__init__()
self.conv_layers = nn.Sequential(
# 第一组卷积层和池化层
nn.Conv2d(3, 64, kernel_size=3, padding=1),
nn.ReLU(inplace=True),
nn.Conv2d(64, 64, kernel_size=3, padding=1),
nn.ReLU(inplace=True),
nn.MaxPool2d(kernel_size=2, stride=2),
# 第二组卷积层和池化层
nn.Conv2d(64, 128, kernel_size=3, padding=1),
nn.ReLU(inplace=True),
nn.Conv2d(128, 128, kernel_size=3, padding=1),
nn.ReLU(inplace=True),
nn.MaxPool2d(kernel_size=2, stride=2),
# 第三组卷积层和池化层
nn.Conv2d(128, 256, kernel_size=3, padding=1),
nn.ReLU(inplace=True),
nn.Conv2d(256, 256, kernel_size=3, padding=1),
nn.ReLU(inplace=True),
nn.Conv2d(256, 256, kernel_size=3, padding=1),
nn.ReLU(inplace=True),
nn.MaxPool2d(kernel_size=2, stride=2),
# 第四组卷积层和池化层
nn.Conv2d(256, 512, kernel_size=3, padding=1),
nn.ReLU(inplace=True),
nn.Conv2d(512, 512, kernel_size=3, padding=1),
nn.ReLU(inplace=True),
nn.Conv2d(512, 512, kernel_size=3, padding=1),
nn.ReLU(inplace=True),
nn.MaxPool2d(kernel_size=2, stride=2),
# 第五组卷积层和池化层
nn.Conv2d(512, 512, kernel_size=3, padding=1),
nn.ReLU(inplace=True),
nn.Conv2d(512, 512, kernel_size=3, padding=1),
nn.ReLU(inplace=True),
nn.Conv2d(512, 512, kernel_size=3, padding=1),
nn.ReLU(inplace=True),
nn.MaxPool2d(kernel_size=2, stride=2),
)
self.fc_layers = nn.Sequential(
# 全连接层
nn.Linear(512 * 7 * 7, 4096),
nn.ReLU(inplace=True),
nn.Dropout(),
nn.Linear(4096, 4096),
nn.ReLU(inplace=True),
nn.Dropout(),
nn.Linear(4096, 10) # 输出层，10个类别的分类
)
def forward(self, x):
x = self.conv_layers(x)
x = x.view(x.size(0), -1)
x = self.fc_layers(x)
return x
相对归一化后较差的运行结果：
epoach=10的运行结果：
C:\Users\Administrator\.conda\envs\pytorch\python.exe "E:\deeplearnig\VGG and Siamese(1)\Train_VGG16.py"
Train Epoch: 1 [0/1400 (0%)] Loss: 2.299627
Train Epoch: 1 [800/1400 (57%)] Loss: 2.314925
Train Epoch: 1 Accuracy: 8.43% Average Loss: 2.323766
Test Accuracy: 10.00% Average Loss: 2.302881
Epoch 1/10, Time: 43.30s
Train Epoch: 2 [0/1400 (0%)] Loss: 2.306078
Train Epoch: 2 [800/1400 (57%)] Loss: 2.261153
Train Epoch: 2 Accuracy: 9.14% Average Loss: 2.304938
Test Accuracy: 10.00% Average Loss: 2.302833
Epoch 2/10, Time: 40.78s
Train Epoch: 3 [0/1400 (0%)] Loss: 2.295484
Train Epoch: 3 [800/1400 (57%)] Loss: 2.291479
Train Epoch: 3 Accuracy: 8.50% Average Loss: 2.305857
Test Accuracy: 10.00% Average Loss: 2.302801
Epoch 3/10, Time: 40.63s
Train Epoch: 4 [0/1400 (0%)] Loss: 2.309008
Train Epoch: 4 [800/1400 (57%)] Loss: 2.304273
Train Epoch: 4 Accuracy: 8.71% Average Loss: 2.304601
Test Accuracy: 10.00% Average Loss: 2.302620
Epoch 4/10, Time: 40.75s
Train Epoch: 5 [0/1400 (0%)] Loss: 2.296353
Train Epoch: 5 [800/1400 (57%)] Loss: 2.309209
Train Epoch: 5 Accuracy: 9.64% Average Loss: 2.304228
Test Accuracy: 10.00% Average Loss: 2.302635
Epoch 5/10, Time: 41.02s
Train Epoch: 6 [0/1400 (0%)] Loss: 2.303827
Train Epoch: 6 [800/1400 (57%)] Loss: 2.305665
Train Epoch: 6 Accuracy: 9.29% Average Loss: 2.303980
Test Accuracy: 10.00% Average Loss: 2.302634
Epoch 6/10, Time: 41.28s
Train Epoch: 7 [0/1400 (0%)] Loss: 2.304016
Train Epoch: 7 [800/1400 (57%)] Loss: 2.300594
Train Epoch: 7 Accuracy: 9.57% Average Loss: 2.304021
Test Accuracy: 10.00% Average Loss: 2.302621
Epoch 7/10, Time: 41.13s
Train Epoch: 8 [0/1400 (0%)] Loss: 2.298562
Train Epoch: 8 [800/1400 (57%)] Loss: 2.306801
Train Epoch: 8 Accuracy: 9.21% Average Loss: 2.303603
Test Accuracy: 10.00% Average Loss: 2.302627
Epoch 8/10, Time: 41.00s
Train Epoch: 9 [0/1400 (0%)] Loss: 2.307916
Train Epoch: 9 [800/1400 (57%)] Loss: 2.303983
Train Epoch: 9 Accuracy: 7.64% Average Loss: 2.303652
Test Accuracy: 10.00% Average Loss: 2.302603
Epoch 9/10, Time: 41.35s
Train Epoch: 10 [0/1400 (0%)] Loss: 2.299228
Train Epoch: 10 [800/1400 (57%)] Loss: 2.303137
Train Epoch: 10 Accuracy: 8.43% Average Loss: 2.303806
Test Accuracy: 10.00% Average Loss: 2.302609
Epoch 10/10, Time: 41.49s
Finished Training
Model saved.
libpng warning: iCCP: known incorrect sRGB profile
libpng warning: iCCP: known incorrect sRGB profile
libpng warning: iCCP: known incorrect sRGB profile
libpng warning: iCCP: known incorrect sRGB profile没有归一化的epoach=100的运行结果：
上述两张图的结果，都是对正则化（归一）未合理使用，导致准确性非常低 ，在构建网络过程中，应注意规避。
总结：使用VGG16网络，注意归一化的使用，忽略归一化会使得训练准确性降低。合理的归一化VGG16网络在100次训练后，在训练集上汽车识别准确率高达95% ，合理开发界面，完成图像分类任务。
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
