Coding-彩色图片分类-基于2D-Unet

编程 评论

摘要

训练一个图片分类神经网络（2D-Unet），包括
1.自定义dataload制作
2,网络定义
3.训练过程
4.测试过程
5.模型评估（准确率）

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
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
import glob  # 导入用于文件路径匹配的模块
from torchvision import transforms  # 导入图像转换模块
from torch.utils import data  # 导入PyTorch数据工具模块
from PIL import Image  # 导入PIL图像处理库
import matplotlib.pyplot as plt  # 导入绘图库
import torch  # 导入PyTorch库
import torch.nn as nn  # 导入PyTorch神经网络模块
import torch.nn.functional as F  # 导入PyTorch函数库
from unet import Unet  # 导入自定义的U-Net模型
import numpy as np  # 导入NumPy库

# 标准化数据
transforms = transforms.Compose([
    transforms.ToTensor(),  # 将图像转换为张量
    transforms.Resize((256, 256)),  # 调整图像大小为256x256
    transforms.Normalize(mean=0.5, std=0.5)  # 标准化图像数据
])

class my_dataset(data.Dataset):
    def __init__(self, imgs_path, annos_path):
        self.imgs_path = imgs_path  # 图像文件路径
        self.annos_path = annos_path  # 标签文件路径

    def __getitem__(self, index):
        img_path = self.imgs_path[index]  # 获取图像路径
        pil_img = Image.open(img_path)  # 使用PIL打开图像
        pil_img = transforms(pil_img)  # 对图像进行预处理

        anno_path = self.annos_path[index]  # 获取标签路径
        anno_img = Image.open(anno_path)  # 使用PIL打开标签图像
        pil_anno = transforms(anno_img)  # 对标签图像进行预处理

        return pil_img, pil_anno

    def __len__(self):
        return len(self.imgs_path)  # 返回数据集的长度

def train(model, train_loader, criterion, optimizer, device):
    model.train()  # 设置模型为训练模式
    train_loss = 0
    for data, label in train_loader:
        data = data.to(device)
        optimizer.zero_grad()  # 清除梯度
        output = model(data)  # 前向传播
        loss = criterion(output, label.to(device).float())  # 计算损失
        loss.backward()  # 反向传播，计算梯度
        optimizer.step()  # 更新模型参数
        train_loss += loss.item() * data.size(0)
          
    train_loss /= len(train_loader.dataset)  # 计算平均训练损失
    return train_loss

def validate(model, val_loader, criterion, device):
    model.eval()  # 设置模型为评估模式
    val_loss = 0
    with torch.no_grad():
        for data, label in val_loader:
            data = data.to(device)
            output = model(data)  # 前向传播
            loss = criterion(output, label.to(device).float())  # 计算损失
            val_loss += loss.item() * data.size(0)
               
    val_loss /= len(val_loader.dataset)  # 计算平均验证损失
    return val_loss 

if __name__ =='__main__':
    # 训练数据集导入
    imgs_path = glob.glob('facade/train_picture/*.png')  # 匹配训练图像文件路径
    label_path = glob.glob('facade/train_label/*.jpg')  # 匹配训练标签文件路径

    # 测试数据集导入
    test_imgs_path = glob.glob('facade/test_picture/*.png')  # 匹配测试图像文件路径
    test_label_path = glob.glob('facade/test_label/*.jpg')  # 匹配测试标签文件路径

    # 对数据和标签排序，确保一一对应
    imgs_path = sorted(imgs_path)
    label_path = sorted(label_path)

    test_imgs_path = sorted(test_imgs_path)
    test_label_path = sorted(test_label_path)

    train_dataset = my_dataset(imgs_path, label_path) 
    test_dataset = my_dataset(test_imgs_path, test_label_path)  # 创建测试数据集对象
    train_loader = data.DataLoader(train_dataset, batch_size=4, shuffle=True)  # 创建训练数据加载器
    test_loader = data.DataLoader(test_dataset, batch_size=4, shuffle=False)  # 创建测试数据加载器

    # 创建U-Net模型
    in_channels = 3  # 输入通道数
    out_channels = 3  # 输出通道数
    model = Unet(in_channels, out_channels)  # 创建U-Net模型对象

    criterion = nn.MSELoss()  # 创建均方误差损失函数对象
    optimizer = torch.optim.Adam(model.parameters(), lr=0.0001)  # 创建优化器对象

    # 将模型和数据移动到GPU上
    device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')  # 检查是否有可用的GPU
    model.to(device)  # 将模型移动到GPU上

    train_losses = []  # 保存训练损失的列表
    val_losses = []  # 保存验证损失的列表
    best_val_loss = np.inf  # 初始化最佳验证损失为正无穷
    best_model = None  # 初始化最佳模型为空
    epoch_times = 300  # 设定迭代次数

    # 训练模型
    for epoch in range(epoch_times):
        train_loss = train(model, train_loader, criterion, optimizer, device)  # 训练模型
        val_loss = validate(model, test_loader, criterion, device)  # 验证模型
        train_losses.append(train_loss)  # 保存训练损失
        val_losses.append(val_loss)  # 保存验证损失

        if val_loss < best_val_loss:
            best_val_loss = val_loss
            best_model = model.state_dict()
            torch.save(best_model, 'ckpt/model.ckpt')  # 保存最佳模型参数
            print("best_val_loss: " + str(val_loss))
            with open("ckpt/model_loss.txt", "w") as f:
                f.write(str(val_loss))

        print('Epoch [{}/{}], Train Loss: {:.4f}, Val Loss: {:.4f}'.format(epoch+1, epoch_times, train_loss, val_loss))

val.py (导入模型进行生成测试)

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
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
from unet import Unet  # 导入自定义的U-Net模型
import torch  # 导入PyTorch库
from torch.utils import data  # 导入PyTorch数据工具模块
from PIL import Image  # 导入PIL图像处理库
from torchvision import transforms  # 导入图像转换模块
import glob  # 导入用于文件路径匹配的模块

# 标准化数据
transforms = transforms.Compose([
    transforms.ToTensor(),  # 将图像转换为张量
    transforms.Resize((256, 256)),  # 调整图像大小为256x256
    transforms.Normalize(mean=0.5, std=0.5)  # 标准化图像数据
])

class my_dataset(data.Dataset):
    def __init__(self, imgs_path, annos_path):
        self.imgs_path = imgs_path  # 图像文件路径
        self.annos_path = annos_path  # 标签文件路径

    def __getitem__(self, index):
        img_path = self.imgs_path[index]  # 获取图像路径
        pil_img = Image.open(img_path)  # 使用PIL打开图像
        pil_img = transforms(pil_img)  # 对图像进行预处理

        anno_path = self.annos_path[index]  # 获取标签路径
        anno_img = Image.open(anno_path)  # 使用PIL打开标签图像
        pil_anno = transforms(anno_img)  # 对标签图像进行预处理

        return pil_img, pil_anno

    def __len__(self):
        return len(self.imgs_path)  # 返回数据集的长度

# 测试数据集导入
test_imgs_path = glob.glob('facade/test_picture/*.png')  # 匹配测试图像文件路径
test_label_path = glob.glob('facade/test_label/*.jpg')  # 匹配测试标签文件路径
test_dataset = my_dataset(test_imgs_path, test_label_path)  # 创建测试数据集对象
test_loader = data.DataLoader(test_dataset, batch_size=1, shuffle=False)  # 创建测试数据加载器

model = Unet(3, 3)  # 创建U-Net模型对象
checkpoint = torch.load('ckpt/model.ckpt')  # 加载模型参数
model.load_state_dict(checkpoint)  # 加载模型参数

for data, label in test_loader:
    data = data * 0.5 + 0.5  # 反标准化图像数据
    output = model(data)  # 前向传播
    output = torch.squeeze(output, 0)  # 去除输出张量的维度为1的维度
    array = output.cpu().detach().numpy().transpose(1, 2, 0)  # 将输出张量转换为NumPy数组，并调整通道顺序为HWC
    image = Image.fromarray((array * 255).astype('uint8'))  # 创建PIL图像对象
    image.save('image1.jpg')  # 保存图像为JPEG文件

unet.py

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
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
import torch
import torch.nn as nn
import torch.nn.functional as F

# UNet的一大层，包含了两层小的卷积
class DoubleConv(nn.Module):
    def __init__(self, in_ch, out_ch):
        super(DoubleConv, self).__init__()
        self.conv = nn.Sequential(
            nn.Conv2d(in_ch, out_ch, 3, padding=1),
            nn.BatchNorm2d(out_ch),
            nn.ReLU(inplace=True),
            nn.Conv2d(out_ch, out_ch, 3, padding=1),
            nn.BatchNorm2d(out_ch),
            nn.ReLU(inplace=True)
        )

    def forward(self, x):
        x = self.conv(x)
        return x

# 定义输入进来的第一层
class InConv(nn.Module):
    def __init__(self, in_ch, out_ch):
        super(InConv, self).__init__()
        self.conv = DoubleConv(in_ch, out_ch)

    def forward(self, x):
        x = self.conv(x)
        return x

# 定义encoder中的向下传播，包括一个maxpool和一大层    
class Down(nn.Module):
    def __init__(self, in_ch, out_ch):
        super(Down, self).__init__()
        self.mpconv = nn.Sequential(
            nn.MaxPool2d(2),
            DoubleConv(in_ch, out_ch)
        )

    def forward(self, x):
        x = self.mpconv(x)
        return x

# 定义decoder中的向上传播
class Up(nn.Module):
    def __init__(self, in_ch, out_ch, bilinear=True):
        super(Up, self).__init__()
        # 定义了self.up的方法
        if bilinear:
            self.up = nn.Upsample(scale_factor=2, mode='bilinear', align_corners=True)
        else:
            self.up = nn.ConvTranspose2d(in_ch // 2, in_ch // 2, 2, stride=2) # // 除以的结果向下取整

        self.conv = DoubleConv(in_ch, out_ch)

    def forward(self, x1, x2):  # x2是左侧的输出，x1是上一大层来的输出
        x1 = self.up(x1)

        diffY = x2.size()[2] - x1.size()[2]
        diffX = x2.size()[3] - x1.size()[3]

        x1 = F.pad(x1, (diffX // 2, diffX - diffX // 2, diffY // 2, diffY - diffY // 2))
        x = torch.cat([x2, x1], dim=1) # 将两个tensor拼接在一起 dim=1：在通道数（C）上进行拼接
        x = self.conv(x)
        return x

# 定义最终的输出
class OutConv(nn.Module):
    def __init__(self, in_ch, out_ch):
        super(OutConv, self).__init__()
        self.conv = nn.Conv2d(in_ch, out_ch, 1)

    def forward(self, x):
        x = self.conv(x)
        return x

class Unet(nn.Module):
    def __init__(self, in_channels, classes): # in_channels 图片的通道数，1为灰度图，3为彩色图
        super(Unet, self).__init__()
        self.n_channels = in_channels
        self.n_classes = classes

        self.inc = InConv(in_channels, 64)
        self.down1 = Down(64, 128)
        self.down2 = Down(128, 256)
        self.down3 = Down(256, 512)
        self.down4 = Down(512, 512)
        self.up1 = Up(1024, 256)
        self.up2 = Up(512, 128)
        self.up3 = Up(256, 64)
        self.up4 = Up(128, 64)
        self.outc = OutConv(64, classes)

    def forward(self, x):
        x1 = self.inc(x)
        x2 = self.down1(x1)
        x3 = self.down2(x2)
        x4 = self.down3(x3)
        x5 = self.down4(x4)
        x = self.up1(x5, x4)
        x = self.up2(x, x3)
        x = self.up3(x, x2)
        x = self.up4(x, x1)
        x = self.outc(x)

        return x

Chandler一头牛马

江上两条红船，寒风斜雨摇摆