卷积神经网络出现的意义在于从人工设计卷积核，到通过样本自动学习卷积核。

LeNet结构

点击查看【processon】

特征图计算公式

卷积计算输出的公式

MINIST入门LeNet网络 - 图1

模型Pytorch代码解析

  1. import torch
  2. import torch.nn as nn
  3. import torch.nn.functional as F
  4. import matplotlib.pyplot as plt
  5. import copy
  6. from torchvision.utils import make_grid
  7. from matplotlib.pyplot import MultipleLocator
  9. class LeNet(nn.Module):
  10.     def __init__(self):
  11.         super(LeNet, self).__init__()
  12.         self.conv1 = nn.Conv2d(1, 6, 5)  # nn.Conv2d(3, 6, 5)
  13.         self.conv2 = nn.Conv2d(6, 16, 5)
  14.         # nn.init()
  15.         self.fc1 = nn.Linear(16 * 4 * 4, 10)
  16.         self.fc2 = nn.Linear(120, 84)
  17.         self.fc3 = nn.Linear(84, 10)
  19.     def forward(self, x):
  20.         #show_graph(x, "origin")
  21.         x = self.conv1(x)
  22.         #show_graph(x, "conv1")
  23.         x = F.relu(x)  # f(x) = max(0, x)
  24.         #show_graph(x, "conv1_relu")
  25.         x = F.max_pool2d(x, 2)
  26.         #show_graph(x, "conv1_relu_maxpool")
  27.         x = self.conv2(x)
  28.         #show_graph(x, "conv2")
  29.         x = F.relu(x)
  30.         #show_graph(x, "conv2_relu")
  31.         x = F.max_pool2d(x, 2)
  32.         #show_graph(x, "conv2_relu_maxpool")
  33.         x = x.view(x.size(0), -1)
  34.         # print(x[0].size())
  35.         x = self.fc1(x)
  36.         # print(x[0].size())
  37.         # x = F.relu(x)
  38.         # x = self.fc2(x)
  39.         # # print(x[0].size())
  40.         # x = F.relu(x)
  41.         # x = self.fc3(x)
  42.         # print(x[0])
  43.         x = F.log_softmax(x, dim=1)
  44.         # print(x[0])
  45.         # exit()
  47.         return x
  50. def show_graph(x, string):
  51.     # y = copy.deepcopy(x[0][0])
  52.     # make_grid(y)
  53.     # plt.imshow(y.cpu().numpy(), cmap='gray')
  54.     # plt.grid()
  55.     # plt.show()
  56.     y = copy.deepcopy(x[0])
  57.     print(y[0])
  58.     y = y * 0.3081 + 0.1307
  59.     y = y.cpu().numpy()
  60.     print(len(y[0]))
  61.     print(y[0])
  62.     ax = plt.gca()
  63.     if len(y[0]) < 10:
  64.         x_major_locator = MultipleLocator(1)
  65.         y_major_locator = MultipleLocator(1)
  66.     else:
  67.         x_major_locator = MultipleLocator(5)
  68.         y_major_locator = MultipleLocator(5)
  69.     ax.xaxis.set_major_locator(x_major_locator)
  70.     ax.yaxis.set_major_locator(y_major_locator)
  71.     for i in range(len(y)):
  72.         plt.imshow(y[0], cmap='gray')
  74.         #plt.grid(b=True, which="major", axis="both", ls="--")
  75.         #plt.xlim(0, len(y[0]))
  76.         #plt.ylim(len(y[0]), 0)
  77.         plt.title("LeNet_{}_{}".format(string, i + 1))
  78.         #plt.savefig("E:/WorkSpace/Pytorch/mnist/model/lenet_feature_map/{}_{}".format(string, i + 1))
  79.         plt.show()

Reference

https://zhuanlan.zhihu.com/p/38200980

https://blog.csdn.net/CSSDCC/article/details/116461271?spm=1001.2014.3001.5501