python使用numpy实现卷积操作
talk is cheap,show you the code
import numpy as np
def Conv2(img, kernel, n, stride):
#img:输入图片;kernel:卷积核值;n:卷积核大小为n*n;stride:步长。
#return:feature map
h, w = img.shape
img = np.pad(img,((1,1),(1,1)),'constant',constant_values=0)
res_h = ((h+2-n)//stride)+1 #卷积边长计算公式:((n+2*p-k)/stride)+1
res_w = ((w+2-n)//stride)+1
res = np.zeros([res_h, res_w])
# print(res)
for i in range(res_h):
for j in range(res_w):
temp = img[i*stride:i*stride+n, j*stride:j*stride+n]
print((i*stride,i*stride+n,j*stride,j*stride+n)) #打印检查卷积核每次卷积的位置对否
temp = np.multiply(kernel, temp)
res[i][j] = temp.sum()
return res
if __name__ == '__main__':
A = np.array([[1, 1, 1, 1, 1], [1, 1, 1, 1, 1], [1, 1, 1, 1, 1], [1, 1, 1, 1, 1]]) # 4行5列
ken = np.array([[2, 2], [2, 2]])
print("result:",Conv2(A, ken, 2, 2))