DAY 43 训练

---

使用 PyTorch 进行猫狗分类与可视化：从零开始的深度学习实践

在当今数字化时代，图像分类任务作为计算机视觉领域的重要基石，正推动着诸多行业的智能化转型。本文将带您深入探索如何利用 PyTorch 框架高效实现猫狗二分类模型的构建与训练，并借助 Grad-CAM 技术直观可视化模型决策依据，助力您快速踏入深度学习实践领域。

数据预处理与加载：构建模型根基

优质的数据预处理是模型成功的关键起点。我们精心设计了针对训练集与测试集的不同预处理流程：

data_transforms = {'train': transforms.Compose([transforms.RandomResizedCrop(224),transforms.RandomHorizontalFlip(),transforms.ToTensor(),transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])]),'test': transforms.Compose([transforms.Resize(256),transforms.CenterCrop(224),transforms.ToTensor(),transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])]),
}

训练集经随机裁剪与水平翻转增强数据多样性，转化为张量后标准化；测试集则仅调整尺寸与裁剪。我们利用 PyTorch 的 ImageFolder 与 DataLoader 简洁加载本地猫狗数据集：

image_datasets = {x: datasets.ImageFolder(data_dir + '/' + x, data_transforms[x]) for x in ['train', 'test']}
dataloaders = {x: DataLoader(image_datasets[x], batch_size=4, shuffle=True, num_workers=0) for x in ['train', 'test']}

模型构建与优化：深度学习核心实现

基于业经验证的 ResNet18 预训练模型，我们巧妙改造以适配猫狗二分类任务：

model = models.resnet18(pretrained=True)
num_ftrs = model.fc.in_features
model.fc = nn.Linear(num_ftrs, 2)  # 猫狗二分类

保留其强大特征提取能力，仅替换全连接层。配备交叉熵损失函数与动量随机梯度下降优化器：

criterion = nn.CrossEntropyLoss()
optimizer = optim.SGD(model.parameters(), lr=0.001, momentum=0.9)

训练与评估：模型性能提升之旅

定义综合训练评估函数，精细把控训练流程：

def train_and_evaluate():device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")model.to(device)num_epochs = 5for epoch in range(num_epochs):print(f'Epoch {epoch}/{num_epochs - 1}')print('-' * 10)for phase in ['train', 'test']:if phase == 'train':model.train()else:model.eval()running_loss = 0.0running_corrects = 0for inputs, labels in dataloaders[phase]:inputs = inputs.to(device)labels = labels.to(device)optimizer.zero_grad()with torch.set_grad_enabled(phase == 'train'):outputs = model(inputs)_, preds = torch.max(outputs, 1)loss = criterion(outputs, labels)if phase == 'train':loss.backward()optimizer.step()running_loss += loss.item()running_corrects += torch.sum(preds == labels.data)epoch_loss = running_loss * inputs.size(0) / dataset_sizes[phase]epoch_acc = running_corrects.double() / dataset_sizes[phase]print(f'{phase} Loss: {epoch_loss:.4f} Acc: {epoch_acc:.4f}')print('Training complete')

多轮迭代训练，动态监测损失与准确率，逐步优化模型性能。

可视化探索：洞察模型决策奥秘

借助 Grad-CAM 技术，我们得以窥探模型决策依据：

def visualize_results():device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")model.eval()cam_extractor = GradCAM(model, 'layer4')inputs, labels = next(iter(dataloaders['test']))input_tensor = inputs[0].unsqueeze(0).to(device)out = model(input_tensor)activation_map = cam_extractor(out.squeeze(0).argmax().item(), out)inv_normalize = transforms.Normalize(mean=[-0.485/0.229, -0.456/0.224, -0.406/0.225],std=[1/0.229, 1/0.224, 1/0.225])img = inv_normalize(input_tensor[0].cpu())img_pil = transforms.ToPILImage()(img)activation_map_pil = transforms.ToPILImage()(torch.from_numpy(activation_map[0].cpu().numpy()))result = overlay_mask(img_pil, activation_map_pil, alpha=0.5)plt.imshow(result)plt.axis('off')plt.show()

精心反归一化还原图像，叠加热力图直观呈现关键区域，助力理解模型决策逻辑。

@浙大疏锦行