# DAY 43 随机函数与广播机制
知识点回顾:
1. 随机张量的生成: torch.randn函数
2. 卷积和池化的计算公式 (可以不掌握, 会自动计算的)
3. pytorch的广播机制: 加法和乘法的广播机制
ps: numpy运算也有类似的广播机制, 基本一致
作业: 自己多举几个例子帮助自己理解即可
# CBAM 注意力
知识点回顾:
1. 通道注意力模块复习
2. 空间注意力模块
3. CBAM的定义
作业：尝试对今天的模型检查参数数目，并用 tensorboard 查看训练过程
@浙大疏锦行
# 导入依赖库 import torch import torch.nn as nn import torch.optim as optim from torch.utils.data import DataLoader from torch.utils.tensorboard import SummaryWriter import torchvision import torchvision.transforms as transforms import matplotlib.pyplot as plt # 1. 模型参数统计工具：统计模型总参数、可训练/不可训练参数数量 def count_model_parameters(model): total_params = 0 trainable_params = 0 non_trainable_params = 0 for param in model.parameters(): param_num = param.numel() total_params += param_num if param.requires_grad: trainable_params += param_num else: non_trainable_params += param_num def format_params(num): if num >= 1e6: return f"{num/1e6:.2f}M" elif num >= 1e3: return f"{num/1e3:.2f}K" else: return f"{num}" print("模型参数统计：") print(f"总参数: {format_params(total_params)} ({total_params:,})") print(f"可训练参数: {format_params(trainable_params)} ({trainable_params:,})") print(f"不可训练参数: {format_params(non_trainable_params)} ({non_trainable_params:,})") return total_params, trainable_params, non_trainable_params # 2. CBAM注意力模块定义 class CBAMBlock(nn.Module): def __init__(self, channels, reduction=16): super().__init__() # 通道注意力 self.avg_pool = nn.AdaptiveAvgPool2d(1) self.max_pool = nn.AdaptiveMaxPool2d(1) self.fc = nn.Sequential( nn.Linear(channels, channels//reduction), nn.ReLU(inplace=True), nn.Linear(channels//reduction, channels) ) # 空间注意力 self.spatial = nn.Sequential( nn.Conv2d(2, 1, kernel_size=7, padding=3, bias=False), nn.Sigmoid() ) self.sigmoid = nn.Sigmoid() def forward(self, x): # 通道注意力计算 avg_out = self.fc(self.avg_pool(x).view(x.size(0), -1)).view(x.size(0), x.size(1), 1, 1) max_out = self.fc(self.max_pool(x).view(x.size(0), -1)).view(x.size(0), x.size(1), 1, 1) channel_att = self.sigmoid(avg_out + max_out) x = x * channel_att # 空间注意力计算 avg_out = torch.mean(x, dim=1, keepdim=True) max_out, _ = torch.max(x, dim=1, keepdim=True) spatial_att = self.spatial(torch.cat([avg_out, max_out], dim=1)) x = x * spatial_att return x # 3. 基于CBAM的CNN模型定义 class CBAM_CNN(nn.Module): def __init__(self, num_classes=10): super().__init__() self.features = nn.Sequential( nn.Conv2d(3, 64, kernel_size=3, padding=1), nn.ReLU(inplace=True), CBAMBlock(64), nn.MaxPool2d(2, 2), nn.Conv2d(64, 128, kernel_size=3, padding=1), nn.ReLU(inplace=True), CBAMBlock(128), nn.MaxPool2d(2, 2), ) self.classifier = nn.Linear(128 * 8 * 8, num_classes) def forward(self, x): x = self.features(x) x = x.view(x.size(0), -1) x = self.classifier(x) return x # 4. 训练函数（集成TensorBoard可视化） def train(model, train_loader, test_loader, criterion, optimizer, scheduler, device, epochs, writer): model.train() all_iter_losses = [] iter_indices = [] for epoch in range(epochs): running_loss = 0.0 correct = 0 total = 0 for batch_idx, (data, target) in enumerate(train_loader): data, target = data.to(device), target.to(device) optimizer.zero_grad() output = model(data) loss = criterion(output, target) loss.backward() optimizer.step() iter_loss = loss.item() global_step = epoch * len(train_loader) + batch_idx + 1 all_iter_losses.append(iter_loss) iter_indices.append(global_step) # TensorBoard记录batch级损失 writer.add_scalar('Train/Batch_Loss', iter_loss, global_step) running_loss += iter_loss _, predicted = output.max(1) total += target.size(0) correct += predicted.eq(target).sum().item() if (batch_idx + 1) % 100 == 0: print(f'Epoch {epoch+1}/{epochs} | Batch {batch_idx+1}/{len(train_loader)} | 单Batch损失: {iter_loss:.4f}') # 记录epoch级训练指标 epoch_train_loss = running_loss / len(train_loader) epoch_train_acc = 100. * correct / total writer.add_scalar('Train/Epoch_Loss', epoch_train_loss, epoch+1) writer.add_scalar('Train/Epoch_Accuracy', epoch_train_acc, epoch+1) # 测试阶段 model.eval() test_loss = 0 correct_test = 0 total_test = 0 with torch.no_grad(): for data, target in test_loader: data, target = data.to(device), target.to(device) output = model(data) test_loss += criterion(output, target).item() _, predicted = output.max(1) total_test += target.size(0) correct_test += predicted.eq(target).sum().item() # 记录epoch级测试指标 epoch_test_loss = test_loss / len(test_loader) epoch_test_acc = 100. * correct_test / total_test writer.add_scalar('Test/Epoch_Loss', epoch_test_loss, epoch+1) writer.add_scalar('Test/Epoch_Accuracy', epoch_test_acc, epoch+1) scheduler.step(epoch_test_loss) print(f'Epoch {epoch+1} 完成 | 训练准确率: {epoch_train_acc:.2f}% | 测试准确率: {epoch_test_acc:.2f}%') writer.close() return epoch_test_acc # 5. 绘图辅助函数 def plot_iter_losses(losses, indices): plt.figure(figsize=(10, 4)) plt.plot(indices, losses, 'b-', alpha=0.7) plt.xlabel('Iteration') plt.ylabel('Loss') plt.title('Training Loss per Iteration') plt.grid(True) plt.show() # 6. 主执行逻辑 if __name__ == "__main__": # 设备配置 device = torch.device('cuda' if torch.cuda.is_available() else 'cpu') # 数据加载（CIFAR10） transform = transforms.Compose([ transforms.RandomHorizontalFlip(), transforms.ToTensor(), transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)) ]) train_dataset = torchvision.datasets.CIFAR10(root='./data', train=True, download=True, transform=transform) test_dataset = torchvision.datasets.CIFAR10(root='./data', train=False, download=True, transform=transform) train_loader = DataLoader(train_dataset, batch_size=64, shuffle=True, num_workers=2) test_loader = DataLoader(test_dataset, batch_size=64, shuffle=False, num_workers=2) # 模型、损失函数、优化器初始化 model = CBAM_CNN().to(device) criterion = nn.CrossEntropyLoss() optimizer = optim.Adam(model.parameters(), lr=0.001) scheduler = optim.lr_scheduler.ReduceLROnPlateau(optimizer, mode='min', patience=3, factor=0.5) # 统计模型参数 count_model_parameters(model) # 初始化TensorBoard writer = SummaryWriter(log_dir='./cbam_logs') # 启动训练 epochs = 50 print("开始模型训练...") final_accuracy = train(model, train_loader, test_loader, criterion, optimizer, scheduler, device, epochs, writer) print(f"训练完成 | 最终测试准确率: {final_accuracy:.2f}%")
