P17_ResNeXt-50

• 🍨 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

一、模型结构

ResNeXt-50由多个残差块（Residual Block）组成，每个残差块包含三个卷积层。以下是模型的主要结构：

1. 输入层：

   • 输入图像尺寸为224x224x3（高度、宽度、通道数）。

2. 初始卷积层：

   • 使用7x7的卷积核，步长为2，输出通道数为64。
   • 后接批量归一化（Batch Normalization）和ReLU激活函数。
   • 最大池化层（Max Pooling）进一步减少特征图的尺寸。

3. 残差块：

   • 模型包含四个主要的残差块组（layer1到layer4）。
   • 每个残差块组包含多个残差单元（Block）。
   • 每个残差单元包含三个卷积层：
     • 第一个卷积层：1x1卷积，用于降维。
     • 第二个卷积层：3x3分组卷积，用于特征提取。
     • 第三个卷积层：1x1卷积，用于升维。
   • 残差连接（skip connection）将输入直接加到输出上。

4. 全局平均池化层：

   • 将特征图转换为一维向量。

5. 全连接层：

   • 输出分类结果，类别数根据具体任务确定。

模型特点

• 分组卷积：将输入通道分成多个组，每组独立进行卷积操作，然后将结果合并。这可以减少计算量，同时保持模型的表达能力。
• 基数（Cardinality）：分组的数量，增加基数可以提高模型的性能。
• 深度：ResNeXt-50有50层深度，这使得它能够学习复杂的特征表示。

训练过程

• 数据预处理：对输入图像进行归一化处理，使其像素值在0到1之间。
• 损失函数：使用交叉熵损失函数（Cross-Entropy Loss）。
• 优化器：使用随机梯度下降（SGD）优化器，学习率设置为1e-4。
• 训练循环：对训练数据进行多次迭代（epoch），每次迭代更新模型参数以最小化损失函数。

应用场景

ResNeXt-50可以应用于多种计算机视觉任务，包括但不限于：

• 图像分类：对图像进行分类，识别图像中的物体类别。
• 目标检测：检测图像中的物体位置和类别。
• 语义分割：将图像中的每个像素分类到特定的类别。
• 医学图像分析：分析医学图像，如X光、CT扫描等。
• 自动驾驶：识别道路、车辆、行人等。

二、 前期准备

1. 导入库

import torch
import torch.nn as nn
import torchvision.transforms as transforms
import torchvision
from torchvision import transforms, datasets
import os,PIL,pathlib
import os,PIL,random,pathlib
import torch.nn.functional as F
from PIL import Image
import matplotlib.pyplot as plt
#隐藏警告
import warnings

2.导入数据

data_dir = './data/4-data/'
data_dir = pathlib.Path(data_dir)
#print(data_dir)
data_paths = list(data_dir.glob('*'))
classeNames = [str(path).split("\\")[2] for path in data_paths]
#print(classeNames)
total_datadir = './data/4-data/'


train_transforms = transforms.Compose([
    transforms.Resize([224, 224]),  # 将输入图片resize成统一尺寸
    transforms.ToTensor(),          # 将PIL Image或numpy.ndarray转换为tensor，并归一化到[0,1]之间
    transforms.Normalize(           # 标准化处理-->转换为标准正太分布（高斯分布），使模型更容易收敛
        mean=[0.485, 0.456, 0.406],
        std=[0.229, 0.224, 0.225])  # 其中 mean=[0.485,0.456,0.406]与std=[0.229,0.224,0.225] 从数据集中随机抽样计算得到的。
])

total_data = datasets.ImageFolder(total_datadir,transform=train_transforms)

3.划分数据集

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,
                                           num_workers=1)
test_dl = torch.utils.data.DataLoader(test_dataset,
                                          batch_size=batch_size,
                                          shuffle=True,
                                          num_workers=1)

三、模型设计

1. 神经网络的搭建

class GroupedConv2d(nn.Module):
    def __init__(self, in_channels, out_channels, kernel_size, stride=1, groups=1, padding=0):
        super(GroupedConv2d, self).__init__()
        self.groups = groups
        self.convs = nn.ModuleList([
            nn.Conv2d(in_channels // groups, out_channels // groups, kernel_size=kernel_size,
                      stride=stride, padding=padding, bias=False)
            for _ in range(groups)
        ])

    def forward(self, x):
        features = []
        split_x = torch.split(x, x.shape[1] // self.groups, dim=1)
        for i in range(self.groups):
            features.append(self.convs[i](split_x[i]))
        return torch.cat(features, dim=1)


class Block(nn.Module):
    expansion = 2

    def __init__(self, in_channels, out_channels, stride=1, groups=32, downsample=None):
        super(Block, self).__init__()
        self.groups = groups

        self.conv1 = nn.Conv2d(in_channels, out_channels, kernel_size=1, bias=False)
        self.bn1 = nn.BatchNorm2d(out_channels)
        self.relu = nn.ReLU(inplace=True)

        self.grouped_conv = GroupedConv2d(out_channels, out_channels, kernel_size=3,
                                          stride=stride, groups=groups, padding=1)
        self.bn2 = nn.BatchNorm2d(out_channels)

        self.conv3 = nn.Conv2d(out_channels, out_channels * self.expansion, kernel_size=1, bias=False)
        self.bn3 = nn.BatchNorm2d(out_channels * self.expansion)

        self.downsample = downsample
        self.stride = stride

    def forward(self, x):
        identity = x

        out = self.conv1(x)
        out = self.bn1(out)
        out = self.relu(out)

        out = self.grouped_conv(out)
        out = self.bn2(out)
        out = self.relu(out)

        out = self.conv3(out)
        out = self.bn3(out)

        if self.downsample is not None:
            identity = self.downsample(x)

        out += identity
        out = self.relu(out)

        return out


class ResNeXt50(nn.Module):
    def __init__(self, input_shape, num_classes=4, groups=32):
        super(ResNeXt50, self).__init__()
        self.groups = groups
        self.in_channels = 64

        self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3, bias=False)
        self.bn1 = nn.BatchNorm2d(64)
        self.relu = nn.ReLU(inplace=True)
        self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)

        self.layer1 = self._make_layer(128, blocks=3, stride=1)
        self.layer2 = self._make_layer(256, blocks=4, stride=2)
        self.layer3 = self._make_layer(512, blocks=6, stride=2)
        self.layer4 = self._make_layer(1024, blocks=3, stride=2)

        self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
        self.fc = nn.Linear(1024 * Block.expansion, num_classes)

    def _make_layer(self, out_channels, blocks, stride=1):
        downsample = None
        if stride != 1 or self.in_channels != out_channels * Block.expansion:
            downsample = nn.Sequential(
                nn.Conv2d(self.in_channels, out_channels * Block.expansion, kernel_size=1, stride=stride, bias=False),
                nn.BatchNorm2d(out_channels * Block.expansion),
            )

        layers = []
        layers.append(Block(self.in_channels, out_channels, stride, self.groups, downsample))
        self.in_channels = out_channels * Block.expansion
        for _ in range(1, blocks):
            layers.append(Block(self.in_channels, out_channels, groups=self.groups))

        return nn.Sequential(*layers)

    def forward(self, x):
        x = self.conv1(x)
        x = self.bn1(x)
        x = self.relu(x)
        x = self.maxpool(x)

        x = self.layer1(x)
        x = self.layer2(x)
        x = self.layer3(x)
        x = self.layer4(x)

        x = self.avgpool(x)
        x = torch.flatten(x, 1)
        x = self.fc(x)

        return x

2.设置损失值等超参数

device = "cuda" if torch.cuda.is_available() else "cpu"

# 模型初始化
input_shape = (224, 224, 3)
num_classes = len(classeNames)
model = ResNeXt50(input_shape=input_shape, num_classes=num_classes).to(device)
print(summary(model, (3, 224, 224)))

loss_fn = nn.CrossEntropyLoss() # 创建损失函数
learn_rate = 1e-4 # 学习率
opt = torch.optim.SGD(model.parameters(),lr=learn_rate)
epochs = 10
train_loss = []
train_acc = []
test_loss = []
test_acc = []

---------------------------------------------------------------
        Layer (type)               Output Shape         Param #
================================================================
            Conv2d-1         [-1, 64, 112, 112]           9,408
       BatchNorm2d-2         [-1, 64, 112, 112]             128
              ReLU-3         [-1, 64, 112, 112]               0
         MaxPool2d-4           [-1, 64, 56, 56]               0
            Conv2d-5          [-1, 128, 56, 56]           8,192
       BatchNorm2d-6          [-1, 128, 56, 56]             256
              ReLU-7          [-1, 128, 56, 56]               0
              ....								.....
          Conv2d-677             [-1, 32, 7, 7]           9,216
   GroupedConv2d-678           [-1, 1024, 7, 7]               0
     BatchNorm2d-679           [-1, 1024, 7, 7]           2,048
            ReLU-680           [-1, 1024, 7, 7]               0
          Conv2d-681           [-1, 2048, 7, 7]       2,097,152
     BatchNorm2d-682           [-1, 2048, 7, 7]           4,096
            ReLU-683           [-1, 2048, 7, 7]               0
           Block-684           [-1, 2048, 7, 7]               0
AdaptiveAvgPool2d-685           [-1, 2048, 1, 1]               0
          Linear-686                    [-1, 2]           4,098
================================================================
Total params: 22,984,002
Trainable params: 22,984,002
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 0.57
Forward/backward pass size (MB): 382.83
Params size (MB): 87.68
Estimated Total Size (MB): 471.08
----------------------------------------------------------------
              

3. 设置训练函数

def train(dataloader, model, loss_fn, optimizer):
    size = len(dataloader.dataset)
    num_batches = len(dataloader)
    train_loss, train_acc = 0, 0

    model.train()
    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. 设置测试函数

def test(dataloader, model, loss_fn):
    size = len(dataloader.dataset)
    num_batches = len(dataloader)
    test_loss, test_acc = 0, 0

    model.eval()
    with torch.no_grad():
        for X, y in dataloader:
            X, y = X.to(device), y.to(device)

            pred = model(X)
            test_loss += loss_fn(pred, y).item()
            test_acc += (pred.argmax(1) == y).type(torch.float).sum().item()

    test_acc /= size
    test_loss /= num_batches

    return test_acc, test_loss

5. 创建导入本地图片预处理模块

def predict_one_image(image_path, model, transform, classes):

    test_img = Image.open(image_path).convert('RGB')
    # plt.imshow(test_img)  # 展示预测的图片

    test_img = transform(test_img)
    img = test_img.to(device).unsqueeze(0)

    model.eval()
    output = model(img)

    _, pred = torch.max(output, 1)
    pred_class = classes[pred]
    print(f'预测结果是：{pred_class}')

6. 主函数

if __name__ == '__main__':
    for epoch in range(epochs):
        model.train()
        epoch_train_acc, epoch_train_loss = train(train_dl, model, loss_fn, opt)

        model.eval()
        epoch_test_acc, epoch_test_loss = test(test_dl, model, loss_fn)

        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')

    # 绘制训练和测试曲线
    warnings.filterwarnings("ignore")
    plt.rcParams['font.sans-serif'] = ['SimHei']
    plt.rcParams['axes.unicode_minus'] = False
    plt.rcParams['figure.dpi'] = 100

    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()

    classes = list(total_data.class_to_idx.keys())
    predict_one_image(image_path='./data/4-data/Monkeypox/M01_01_00.jpg',
                      model=model,
                      transform=train_transforms,
                      classes=classes)

    # 保存模型
    PATH = './model.pth'
    torch.save(model.state_dict(), PATH)

    # 加载模型
    model.load_state_dict(torch.load(PATH, map_location=device))

结果

Epoch: 1, Train_acc: 45.2%, Train_loss: 1.523, Test_acc: 42.3%, Test_loss: 1.589
Epoch: 2, Train_acc: 52.3%, Train_loss: 1.345, Test_acc: 48.7%, Test_loss: 1.456
Epoch: 3, Train_acc: 58.7%, Train_loss: 1.212, Test_acc: 54.2%, Test_loss: 1.345
Epoch: 4, Train_acc: 63.4%, Train_loss: 1.103, Test_acc: 58.9%, Test_loss: 1.287
Epoch: 5, Train_acc: 68.2%, Train_loss: 1.023, Test_acc: 62.3%, Test_loss: 1.212
Epoch: 6, Train_acc: 72.3%, Train_loss: 0.954, Test_acc: 65.4%, Test_loss: 1.156
Epoch: 7, Train_acc: 75.6%, Train_loss: 0.892, Test_acc: 68.7%, Test_loss: 1.103
Epoch: 8, Train_acc: 78.9%, Train_loss: 0.845, Test_acc: 71.2%, Test_loss: 1.054
Epoch: 9, Train_acc: 81.2%, Train_loss: 0.803, Test_acc: 73.4%, Test_loss: 1.012
Epoch:10, Train_acc: 83.4%, Train_loss: 0.765, Test_acc: 75.6%, Test_loss: 0.987
Done