21卷积层中的多输入与多输出通道.zip

import torch ''' 实现多输入通道的互相关运算 ''' print('实现多输入通道的互相关运算') def corr2d(X,K): #计算二维互相关运算 h , w = K.shape # 表示二维卷积核的 长 和 宽 Y = torch.zeros((X.shape[0] - h + 1 , X.shape[1] - w + 1)) for i in range(Y.shape[0]): for j in range(Y.shape[1]): Y[i,j] = (X[i:i + h,j:j + w] * K).sum() return Y def corr2d_multi_in(X,K): return sum(corr2d(x,k) for x,k in zip(X,K)) X = torch.tensor([[[0.0,1.0,2.0],[3.0,4.0,5.0],[6.0,7.0,8.0]],[[1.0,2.0,3.0],[4.0,5.0,6.0],[7.0,8.0,9.0]]]) print(X) K = torch.tensor([[[0.0,1.0],[2.0,3.0]],[[1.0,2.0],[3.0,4.0]]]) print(K) corr2d_multi_in(X, K) print(corr2d_multi_in(X,K)) print('*'*100) ''' 实现多输出通道的运算 ''' print('实现多输出通道的运算') def corr2d_multi_out(X,K): return torch.stack([corr2d_multi_in(X,k) for k in K],0) K = torch.stack((K,K+1,K+2),0) ##使用3维的卷积核 print(K) print(corr2d_multi_out(X,K)) print('*'*100) ''' 验证1*1的卷积核等价于全连接网络 ''' print('1X1卷积') def corr2d_multi_in_out_1X1(X,K): c_i , h , w = X.shape c_o = K.shape[0] X = X.reshape((c_i , h * w)) K = K.reshape((c_o,c_i)) Y = torch.matmul(K,X) return Y.reshape((c_o,h,w)) X1 = torch.normal(0,1,(3,3,3)) K1 = torch.normal(0,1,(2,3,1,1)) Y1 = corr2d_multi_in_out_1X1(X1,K1) Y2 = corr2d_multi_out(X1,K1) print('Y1',Y1) print('Y2',Y2) assert float(torch.abs(Y1 - Y2).sum()) < 1e-6