import torch
import torch.nn as nn
from PIL import Image
import numpy as np
import matplotlib.pyplot as plt
from torchvision import transforms# 1. 數據預處理：加載并轉換圖像
def load_image(image_path):"""將圖像轉為PyTorch張量，添加批次維度"""# 預處理流程：調整尺寸→轉Tensor→標準化（便于網絡訓練）transform = transforms.Compose([transforms.Resize((224, 224)),  # 統一尺寸為224×224transforms.ToTensor(),          # 轉為Tensor，形狀[C, H, W]，值范圍[0,1]transforms.Normalize(mean=[0.5, 0.5, 0.5], std=[0.5, 0.5, 0.5])  # 標準化到[-1,1]])image = Image.open(image_path).convert('RGB')  # 打開圖像并轉為RGB三通道image_tensor = transform(image).unsqueeze(0)   # 添加批次維度，形狀變為[1, 3, 224, 224]return image, image_tensor# 2. 定義卷積網絡（含2個卷積層）
class ConvDemo(nn.Module):def __init__(self):super(ConvDemo, self).__init__()# 第1個卷積層：保持尺寸不變self.conv1 = nn.Conv2d(in_channels=3,         # 輸入：RGB三通道out_channels=16,       # 輸出：16個特征圖（16個卷積核）kernel_size=3,         # 3×3卷積核stride=1,              # 步長1padding=1              # padding=1：保持輸出尺寸=輸入尺寸)# 第2個卷積層：尺寸減半self.conv2 = nn.Conv2d(in_channels=16,        # 輸入：上一層的16個通道out_channels=8,        # 輸出：8個特征圖kernel_size=5,         # 5×5卷積核stride=2,              # 步長2：輸出尺寸減半padding=2              # padding=2：配合步長2保持尺寸比例)def forward(self, x):x1 = self.conv1(x)  # 經過第1層卷積的輸出x2 = self.conv2(x1) # 經過第2層卷積的輸出return x1, x2# 3. 可視化函數：展示卷積核與特征圖
def visualize_results(original_image, conv1_output, conv2_output, conv1_kernels):plt.figure(figsize=(16, 12))  # 設置畫布大小# 子圖1：原始圖像plt.subplot(3, 1, 1)plt.title('Original Image')plt.imshow(original_image)plt.axis('off')  # 關閉坐標軸# 子圖2：第1層卷積核（前4個）plt.subplot(3, 2, 3)plt.title('First 4 Kernels of Conv1 (3×3)')kernels = conv1_kernels.detach().numpy()  # 轉為numpy數組（脫離計算圖）kernel_grid = np.zeros((3*3, 3*4))  # 構建3×3核的展示網格（4個核）for i in range(4):kernel = kernels[i].transpose(1, 2, 0)  # 轉為[H, W, C]格式（便于顯示）kernel = (kernel - kernel.min()) / (kernel.max() - kernel.min())  # 歸一化到[0,1]kernel_grid[:, i*3:(i+1)*3] = kernel.reshape(9, 3)  # 排列到網格plt.imshow(kernel_grid, cmap='gray')plt.axis('off')# 子圖3：第1層特征圖（前4個）plt.subplot(3, 2, 4)plt.title('First 4 Feature Maps of Conv1')feat1 = conv1_output[0].detach().numpy()  # 取第1個樣本的特征圖feat1_grid = np.zeros((224, 224*4))  # 構建特征圖展示網格for i in range(4):feat = feat1[i]feat = (feat - feat.min()) / (feat.max() - feat.min())  # 歸一化feat1_grid[:, i*224:(i+1)*224] = feat  # 排列到網格plt.imshow(feat1_grid, cmap='gray')plt.axis('off')# 子圖4：第2層特征圖（前4個）plt.subplot(3, 1, 3)plt.title('First 4 Feature Maps of Conv2 (Downsampled)')feat2 = conv2_output[0].detach().numpy()feat2_grid = np.zeros((112, 112*4))  # 步長2，尺寸變為112×112for i in range(4):feat = feat2[i]feat = (feat - feat.min()) / (feat.max() - feat.min())  # 歸一化feat2_grid[:, i*112:(i+1)*112] = feat  # 排列到網格plt.imshow(feat2_grid, cmap='gray')plt.axis('off')plt.tight_layout()  # 自動調整子圖間距plt.show()# 4. 主函數：執行流程
if __name__ == '__main__':# 加載圖像（替換為你的圖像路徑，如'cat.jpg'）image, image_tensor = load_image('example.jpg')print(f"輸入圖像形狀: {image_tensor.shape}")  # 輸出：[1, 3, 224, 224]# 初始化模型并執行前向傳播model = ConvDemo()conv1_out, conv2_out = model(image_tensor)  # 得到兩層卷積的輸出# 打印輸出形狀（驗證尺寸計算）print(f"Conv1輸出形狀: {conv1_out.shape}")  # 輸出：[1, 16, 224, 224]（尺寸不變）print(f"Conv2輸出形狀: {conv2_out.shape}")  # 輸出：[1, 8, 112, 112]（尺寸減半）# 獲取第1層卷積核（權重參數）conv1_kernels = model.conv1.weight  # 形狀：[16, 3, 3, 3]（16個3×3×3的核）# 可視化結果visualize_results(original_image=image,conv1_output=conv1_out,conv2_output=conv2_out,conv1_kernels=conv1_kernels)