PyTorch在CIFAR-10数据集上的训练及测试过程

发表于 更新于 分类于
用Pytorch在CIFAR-10数据集上测试，定义简单两层卷积、实现可视化和用自取图像测试。

PyTorch在CIFAR-10数据集上的训练及测试过程

package Version

  • python 3.6.5
  • pytorch 0.4.1

其他相关：

CIFAR-10数据集

PyTorch读取Cifar数据集并显示图片

卷积神经网络中的参数计算

1
2
3
4
5
6
7
8
9
10
11
12
13
14
# 导入包
# basic
import matplotlib.pyplot as plt # plt 用于显示图片
import matplotlib.image as mpimg # mpimg 用于读取图片
import numpy as np

#resize功能
from scipy import misc

# pytorch
import torch 
import torch.nn as nn
import torchvision
import torchvision.transforms as transforms
1
2
3
# 设备设置
torch.cuda.set_device(1) # 这句用来设置pytorch在哪块GPU上运行，pytorch-cpu版本不需要运行这句
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
1
2
3
4
5
# 超参数设置
num_epochs = 5
num_classes = 10
batch_size = 32
learning_rate = 0.001

加载CIFAR-10数据集

  • 加载数据集
  • 数据增广
1
2
3
# cifar10 分类索引
classes = ('plane', 'car', 'bird', 'cat',
           'deer', 'dog', 'frog', 'horse', 'ship', 'truck')
1
2
3
4
5
6
7
8
9
10
11
12
13
14
# 数据增广方法
transform = transforms.Compose([
    # +4填充至36x36
    transforms.Pad(4),
    # 随机水平翻转
    transforms.RandomHorizontalFlip(), 
    # 随机裁剪至32x32
    transforms.RandomCrop(32), 
    # 转换至Tensor
    transforms.ToTensor(),
    #  归一化
#     transforms.Normalize(mean=(0.5, 0.5, 0.5),   # 3 for RGB channels
#                                      std=(0.5, 0.5, 0.5))
    ])
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
# cifar10路径
cifar10Path = './cifar'

#  训练数据集
train_dataset = torchvision.datasets.CIFAR10(root=cifar10Path,
                                             train=True, 
                                             transform=transform,
                                             download=True)

# 测试数据集
test_dataset = torchvision.datasets.CIFAR10(root=cifar10Path,
                                            train=False, 
                                            transform=transform)

# 生成数据加载器
# 训练数据加载器
train_loader = torch.utils.data.DataLoader(dataset=train_dataset,
                                           batch_size=batch_size, 
                                           shuffle=True)
# 测试数据加载器
test_loader = torch.utils.data.DataLoader(dataset=test_dataset,
                                          batch_size=batch_size, 
                                          shuffle=False)
Files already downloaded and verified

1
2
3
4
# 查看数据,取一组batch
data_iter = iter(test_loader)

images, labels = next(data_iter)
1
2
3
4
5
6
# 取batch中的一张图像
idx = 15
image = images[idx].numpy()
image = np.transpose(image, (1,2,0))
plt.imshow(image)
classes[labels[idx].numpy()]
'ship'

png

网络模型设计和训练：自定义卷积神经网络

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
# 搭建卷积神经网络模型
# 两个卷积层
class ConvNet(nn.Module):
    def __init__(self, num_classes=10):
        super(ConvNet, self).__init__()
        self.conv1 = nn.Sequential(
            # 卷积层计算
            nn.Conv2d(3, 16, kernel_size=5, stride=1, padding=2),
            #  批归一化
            nn.BatchNorm2d(16),
            #ReLU激活函数
            nn.ReLU(),
            # 池化层：最大池化
            nn.MaxPool2d(kernel_size=2, stride=2))
        
        self.conv2 = nn.Sequential(
            nn.Conv2d(16, 32, kernel_size=5, stride=1, padding=2),
            nn.BatchNorm2d(32),
            nn.ReLU(),
            nn.MaxPool2d(kernel_size=2, stride=2))
        
        self.fc = nn.Linear(8*8*32, num_classes)
        
    # 定义前向传播顺序
    def forward(self, x):
        out = self.conv1(x)
        out = self.conv2(out)
        out = out.reshape(out.size(0), -1)
        out = self.fc(out)
        return out
1
2
# 实例化一个模型，并迁移至gpu
model = ConvNet(num_classes).to(device)
1
2
3
# 定义损失函数和优化器
criterion = nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters(), lr=learning_rate)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
total_step = len(train_loader)
for epoch in range(num_epochs):
    for i, (images, labels) in enumerate(train_loader):
        # 注意模型在GPU中，数据也要搬到GPU中
        images = images.to(device)
        labels = labels.to(device)
        
        # 前向传播
        outputs = model(images)
        loss = criterion(outputs, labels)
        
        # 反向传播和优化
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()
        
        if (i+1) % 100 == 0:
            print ('Epoch [{}/{}], Step [{}/{}], Loss: {:.4f}' 
                   .format(epoch+1, num_epochs, i+1, total_step, loss.item()))
Epoch [1/5], Step [100/1563], Loss: 1.9562
Epoch [1/5], Step [200/1563], Loss: 1.6313
Epoch [1/5], Step [300/1563], Loss: 1.6405
Epoch [1/5], Step [400/1563], Loss: 1.7303
Epoch [1/5], Step [500/1563], Loss: 1.4697
Epoch [1/5], Step [600/1563], Loss: 1.5183
Epoch [1/5], Step [700/1563], Loss: 1.3547

...

Epoch [5/5], Step [600/1563], Loss: 1.1019
Epoch [5/5], Step [700/1563], Loss: 0.8237
Epoch [5/5], Step [800/1563], Loss: 0.8497
Epoch [5/5], Step [900/1563], Loss: 0.7738
Epoch [5/5], Step [1000/1563], Loss: 0.7937
Epoch [5/5], Step [1100/1563], Loss: 1.1982
Epoch [5/5], Step [1200/1563], Loss: 1.1113
Epoch [5/5], Step [1300/1563], Loss: 0.9233
Epoch [5/5], Step [1400/1563], Loss: 1.3893
Epoch [5/5], Step [1500/1563], Loss: 0.9238

模型测试和保存

1
2
# 设置为评估模式
model.eval()
ConvNet(
  (conv1): Sequential(
    (0): Conv2d(3, 16, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
    (1): BatchNorm2d(16, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (2): ReLU()
    (3): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  )
  (conv2): Sequential(
    (0): Conv2d(16, 32, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2))
    (1): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (2): ReLU()
    (3): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  )
  (fc): Linear(in_features=2048, out_features=10, bias=True)
)

1
2
3
4
5
6
7
8
9
10
11
12
13
# 节省计算资源，不去计算梯度
with torch.no_grad():
    correct = 0
    total = 0
    for images, labels in test_loader:
        images = images.to(device)
        labels = labels.to(device)
        outputs = model(images)
        _, predicted = torch.max(outputs.data, 1)
        total += labels.size(0)
        correct += (predicted == labels).sum().item()

    print('Test Accuracy of the model on the test images: {} %'.format(100 * correct / total))
Test Accuracy of the model on the test images: 59.35 %

准确率是低了点，因为模型太简单了。

1
2
#  保存模型
torch.save(model.state_dict(), 'model.ckpt')

可视化测试（数据集图像和其他图像测试）

取测试集图像测试

1
2
3
# 查看数据,取一组batch
data_iter = iter(test_loader)
images, labels = next(data_iter)
1
2
3
4
5
6
# 取batch中的一张图像，显示图像和真实标签
idx = 10
image = images[idx].numpy()
image = np.transpose(image, (1,2,0))
plt.imshow(image)
classes[labels[idx].numpy()]
'plane'

png

放入模型进行测试

1
2
3
4
5
# 转换为（B,C,H,W）大小
imagebatch = image.reshape(-1,3,32,32)

# 转换为torch tensor
image_tensor = torch.from_numpy(imagebatch)
1
2
3
4
5
6
7
# 调用模型进行评估
model.eval() 
output = model(image_tensor.to(device))
_, predicted = torch.max(output.data, 1)
pre = predicted.cpu().numpy()
print(pre) # 查看预测结果ID
print(classes[pre[0]])
[0]
plane

注意，准确率还很低，预测出错没事情。后面改善网络结构就好了。

读入一张自己图像进行测试

1
2
3
4
5
6
7
8
9
10
# 读取图像
srcPath = 'horse.jpg'
src = mpimg.imread(srcPath)# 读取和代码处于同一目录下的 图片
# 此时 lena 就已经是一个 np.array 了，可以对它进行任意处理
# 原图大小
print(src.shape) 

plt.imshow(src) # 显示图片
plt.axis('off') # 不显示坐标轴
plt.show()
(193, 260, 3)

png

1
2
3
4
5
# resize至32x32
newImg = misc.imresize(src,(32,32))
plt.imshow(newImg) # 显示图片
plt.axis('off') # 不显示坐标轴
plt.show()
/home/ubuntu/anaconda3/lib/python3.6/site-packages/ipykernel_launcher.py:2: DeprecationWarning: `imresize` is deprecated!
`imresize` is deprecated in SciPy 1.0.0, and will be removed in 1.2.0.
Use ``skimage.transform.resize`` instead.

png

1
2
3
4
5
# 转换为（B,C,H,W）大小
imagebatch = newImg.reshape(-1,3,32,32)

# 转换为torch tensor
image_tensor = torch.from_numpy(imagebatch).float()
1
2
3
4
5
6
7
# 调用模型进行评估
model.eval() 
output = model(image_tensor.to(device))
_, predicted = torch.max(output.data, 1)
pre = predicted.cpu().numpy()
print(pre) # 查看预测结果ID
print(classes[pre[0]])
[1]
car

还是预测出错，但是过程就是这么个过程。后续可以通过改善模型来提高准确率。