输出是一个单通道,因为不管输入有多少个通道,最后都会把结果相加
因为1x1卷积层每次只识别一个像素,而不查看该像素与周围像素的关系,所以它并不识别空间信息
import torch
from d2l import torch as d2l
# 多通道互相关运算
def corr2d_multi_in(X, K):
return sum(d2l.corr2d(x, k) for x, k in zip(X, K))
# 验证互相关运算的输出
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]]])
K = torch.tensor([[[0.0, 1.0], [2.0, 3.0]], [[1.0, 2.0], [3.0, 4.0]]])
print(corr2d_multi_in(X, K))
# 多个通道的输出的互相关函数
def corr2d_multi_in_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)
print(K.shape)
print(corr2d_multi_in_out(X, K))
# 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))
X = torch.normal(0, 1, (3, 3, 3))
K = torch.normal(0, 1, (2, 3, 1, 1))
Y1 = corr2d_multi_in_out_1x1(X, K)
Y2 = corr2d_multi_in_out(X, K)
assert float(torch.abs(Y1 - Y2).sum()) < 1e-6
输入的高宽都减半的情况下,通常会把输出的通道数加一倍。(空间信息压缩了,把提取出来的信息保存在更多的通道里)