1. 卷积

• 输入A张量为：
  $$\begin{array}{c}A=\left[\begin{array}{cccc}0& 1& 2& 3\\ & & & \\ 4& 5& 6& 7\\ & & & \\ 8& 9& 10& 11\\ & & & \\ 12& 13& 14& 15\end{array}\right]\end{array}$$
• 卷积核矩阵
  $$\begin{array}{c}weight=\left[\begin{array}{ccc}0& 1& 2\\ & & \\ 3& 4& 5\\ & & \\ 6& 7& 8\end{array}\right],bias=10\end{array}$$
  

2. python 代码

• pytorch

import torch
import torch.nn as nn
import torch.nn.functional as Ftorch.set_printoptions(precision=3, sci_mode=False)if __name__ == "__main__":run_code = 0batch_size = 1in_channels = 1out_channels = 1kernel_size = 4input_h = 4input_w = 4input_total = input_h * input_winput_matrix = torch.arange(input_total).reshape(batch_size, in_channels, input_h, input_w).to(torch.float)my_conv2d = nn.Conv2d(in_channels=in_channels, out_channels=out_channels, kernel_size=kernel_size, bias=True,stride=1)print(f"my_conv2d={my_conv2d}")my_conv2d_weight = torch.arange(9).reshape((1, 1, 3, 3)).to(torch.float)my_conv2d_bias = torch.tensor([10.0])my_conv2d.weight = nn.Parameter(my_conv2d_weight)my_conv2d.bias = nn.Parameter(my_conv2d_bias)print(f"my_conv2d_weight=\n{my_conv2d_weight}")print(f"my_conv2d_bias=\n{my_conv2d_bias}")output_matrix = my_conv2d(input_matrix)print(f"input_matrix=\n{input_matrix}")print(f"output_matrix=\n{output_matrix}")output_matrix_F = F.conv2d(input=input_matrix, weight=my_conv2d_weight, bias=my_conv2d_bias)print(f"output_matrix_F=\n{output_matrix_F}")

• 结果：

my_conv2d=Conv2d(1, 1, kernel_size=(4, 4), stride=(1, 1))
my_conv2d_weight=
tensor([[[[0., 1., 2.],[3., 4., 5.],[6., 7., 8.]]]])
my_conv2d_bias=
tensor([10.])
input_matrix=
tensor([[[[ 0.,  1.,  2.,  3.],[ 4.,  5.,  6.,  7.],[ 8.,  9., 10., 11.],[12., 13., 14., 15.]]]])
output_matrix=
tensor([[[[268., 304.],[412., 448.]]]], grad_fn=<ConvolutionBackward0>)
output_matrix_F=
tensor([[[[268., 304.],[412., 448.]]]])

3. notes