day43 CNN及Grad-CAM实战——以Intel Image Classification为例

步骤一：数据准备

📌 基本信息
名称：Intel Image Classification Dataset

来源：由 Intel 提供，广泛用于图像分类任务入门（原始来源：Kaggle）

任务类型：多类别图像分类（多类场景识别）

图像数量：25,000+ 张图像

图像尺寸：150×150 彩色 JPG 图像

类别数量：6 类

🏷️ 类别（共 6 类）
每一张图片都属于以下六类之一，每类对应一个子文件夹：

类别名称 内容描述
buildings 各种建筑场景（如城市楼宇）
forest 林地、森林风景图
glacier 冰川、雪地场景
mountain 山地或丘陵环境
sea 海洋、水体场景
street 城市街景、人行道、公路等

这些类别大多属于自然或人造景观，非常适合做场景识别模型实验。

📁 数据结构
通常包含以下三个子文件夹（可能因版本不同略有差异）：

Intel Image Classification/
├── seg_train/ ← 训练集（每类一个子文件夹）
├── seg_test/ ← 测试集（结构同上）
├── seg_pred/ 或 prediction/ ← 预测集（未标注图像，用于模型测试）

子结构示例（以训练集为例）：

seg_train/
├── buildings/
│ ├── img_1.jpg
│ └── …
├── forest/
├── glacier/
├── mountain/
├── sea/
└── street/
📊 图像数量示例（以 Kaggle 数据集为例）：
数据集部分
训练集 2,000 张
测试集 500 张
预测集 无标签

from torchvision import datasets, transforms
from torch.utils.data import DataLoader# 路径适配
train_dir = "./data/Intel Image Classification/seg_train"
test_dir = "./data/Intel Image Classification/seg_test"
# 图像预处理
transform = transforms.Compose([transforms.Resize((224, 224)),transforms.ToTensor(),
])# 加载训练与测试集
train_dataset = datasets.ImageFolder(train_dir, transform=transform)
test_dataset = datasets.ImageFolder(test_dir, transform=transform)train_loader = DataLoader(train_dataset, batch_size=32, shuffle=True)
test_loader = DataLoader(test_dataset, batch_size=32)

CNN训练代码(此处以resnet18为例)

import torch
import torch.nn as nn
from torchvision import modelsdevice = torch.device("cuda" if torch.cuda.is_available() else "cpu")model = models.resnet18(pretrained=True)
model.fc = nn.Linear(model.fc.in_features, len(train_dataset.classes))
model = model.to(device)criterion = nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters(), lr=1e-4)# 训练循环
for epoch in range(5):model.train()running_loss = 0for imgs, labels in train_loader:imgs, labels = imgs.to(device), labels.to(device)optimizer.zero_grad()outputs = model(imgs)loss = criterion(outputs, labels)loss.backward()optimizer.step()running_loss += loss.item()print(f"Epoch {epoch+1}, Loss: {running_loss / len(train_loader):.4f}")

d:\Anaconda\envs\DL\lib\site-packages\torchvision\models\_utils.py:208: UserWarning: The parameter 'pretrained' is deprecated since 0.13 and may be removed in the future, please use 'weights' instead.warnings.warn(
d:\Anaconda\envs\DL\lib\site-packages\torchvision\models\_utils.py:223: UserWarning: Arguments other than a weight enum or `None` for 'weights' are deprecated since 0.13 and may be removed in the future. The current behavior is equivalent to passing `weights=ResNet18_Weights.IMAGENET1K_V1`. You can also use `weights=ResNet18_Weights.DEFAULT` to get the most up-to-date weights.warnings.warn(msg)
Downloading: "https://download.pytorch.org/models/resnet18-f37072fd.pth" to C:\Users\Administrator/.cache\torch\hub\checkpoints\resnet18-f37072fd.pth
100%|██████████| 44.7M/44.7M [00:04<00:00, 10.8MB/s]Epoch 1, Loss: 0.2814
Epoch 2, Loss: 0.1280
Epoch 3, Loss: 0.0626
Epoch 4, Loss: 0.0495
Epoch 5, Loss: 0.0345

Grad-CAM可视化（可视化 /data/Intel Image Classification/seg_test/buildings 中每个类别的随机任一图片）

import os
import random
from PIL import Image
import numpy as np
import matplotlib.pyplot as plt
import torch
from torchvision import transforms
from pytorch_grad_cam import GradCAM
from pytorch_grad_cam.utils.model_targets import ClassifierOutputTarget
from pytorch_grad_cam.utils.image import show_cam_on_imagedef show_gradcam_side_by_side_classes(model,folder_path,target_layers,class_names=None,device="cuda" if torch.cuda.is_available() else "cpu"
):model.eval()transform = transforms.Compose([transforms.Resize((224, 224)),transforms.ToTensor()])all_classes = class_names if class_names else os.listdir(folder_path)cam = GradCAM(model=model, target_layers=target_layers)for cls in all_classes:cls_path = os.path.join(folder_path, cls)image_files = [f for f in os.listdir(cls_path) if f.lower().endswith((".jpg", ".jpeg", ".png"))]img_file = random.choice(image_files)img_path = os.path.join(cls_path, img_file)print(f"\n▶️ 类别: {cls} | 图像: {img_file}")image = Image.open(img_path).convert("RGB")input_tensor = transform(image).unsqueeze(0).to(device)output = model(input_tensor)pred_class = output.argmax().item()grayscale_cam = cam(input_tensor=input_tensor, targets=[ClassifierOutputTarget(pred_class)])grayscale_cam = grayscale_cam[0, :]rgb_img = np.array(image.resize((224, 224))) / 255.0visualization = show_cam_on_image(rgb_img, grayscale_cam, use_rgb=True)# 横向并排显示fig, axes = plt.subplots(1, 2, figsize=(10, 4))axes[0].imshow(rgb_img)axes[0].set_title(f"[原图] 类别: {cls}")axes[0].axis('off')axes[1].imshow(visualization)axes[1].set_title(f"[Grad-CAM] 预测: {class_names[pred_class]}")axes[1].axis('off')plt.tight_layout()plt.show()show_gradcam_side_by_side_classes(model=model,folder_path=r".\data\Intel Image Classification\seg_test",target_layers=[model.layer4[-1]],class_names=train_dataset.classes
)

▶️ 类别: buildings | 图像: 21496.jpg

▶️ 类别: forest | 图像: 20328.jpg

▶️ 类别: glacier | 图像: 23448.jpg

▶️ 类别: mountain | 图像: 22117.jpg

▶️ 类别: sea | 图像: 20172.jpg

▶️ 类别: street | 图像: 20572.jpg