@浙大疏锦行
import torch import torch.nn as nn import torch.nn.functional as F class SimpleCNN(nn.Module): def __init__(self): super(SimpleCNN, self).__init__() # 1. 卷积层 (Convolutional Layer) # 输入: 1个通道(灰度图), 输出: 32个特征图, 卷积核: 3x3 self.conv1 = nn.Conv2d(in_channels=1, out_channels=32, kernel_size=3, padding=1) # 2. 池化层 (Pooling Layer) # 2x2 最大池化，将图片尺寸减半 self.pool = nn.MaxPool2d(kernel_size=2, stride=2) # 3. 全连接层 (Fully Connected Layer) # 假设输入图片是 28x28: # 经过 conv1 (padding=1) -> 28x28 # 经过 pool -> 14x14 # 所以展平后的维度是: 32(通道) * 14 * 14 self.fc1 = nn.Linear(32 * 14 * 14, 10) # 输出 10 个类别 (例如数字 0-9) def forward(self, x): # 前向传播逻辑 x = self.conv1(x) x = F.relu(x) # 激活函数 x = self.pool(x) # 下采样 x = x.view(-1, 32 * 14 * 14) # 展平 (Flatten) x = self.fc1(x) return x # --- 测试代码 --- # 创建模型实例 model = SimpleCNN() print(model) # 模拟一张随机生成的图片 (Batch_size=1, Channel=1, Height=28, Width=28) dummy_input = torch.randn(1, 1, 28, 28) output = model(dummy_input) print(f"\n输出形状: {output.shape}") # 应该是 [1, 10] import tensorflow as tf from tensorflow.keras import layers, models def create_simple_cnn(): model = models.Sequential() # 1. 卷积层 + 激活函数 # input_shape=(28, 28, 1) 对应 高, 宽, 通道数 model.add(layers.Conv2D(32, (3, 3), activation='relu', input_shape=(28, 28, 1))) # 2. 池化层 model.add(layers.MaxPooling2D((2, 2))) # 3. 展平层 (把多维特征图变成一维向量) model.add(layers.Flatten()) # 4. 全连接层 (输出层) model.add(layers.Dense(10, activation='softmax')) # 10个类别 return model # --- 测试代码 --- model = create_simple_cnn() # 打印模型结构概览 model.summary()
