winograd算法python小白教程

mathilde27

于 2019-12-18 08:46:09 发布

阅读量824

点赞数 1

本文链接：https://blog.youkuaiyun.com/Chunying27/article/details/103591157

版权

This tutorial shows how to compute wino_f23:

define transform matrix

G_F23 = np.array([
     [ 1.0,  0.0, 0.0 ],
     [ 0.5,  0.5, 0.5 ],
     [ 0.5, -0.5, 0.5 ],
     [ 0.0,  0.0, 1.0 ]])
 Bt_F23 = np.array([
     [ 1.0,  0.0, -1.0,  0.0 ],
     [ 0.0,  1.0,  1.0,  0.0 ],
     [ 0.0, -1.0,  1.0,  0.0 ],
     [ 0.0,  1.0,  0.0, -1.0 ]])
 At_F23 = np.array([
     [ 1.0, 1.0,  1.0,  0.0 ],
     [ 0.0, 1.0, -1.0, -1.0 ]])

compute transformation for input, kernel, output

 def trans_kernel(g):
     return np.dot(np.dot(G_F23,g),G_F23.T)
 def trans_input(d):
     return np.dot(np.dot(Bt_F23,d),Bt_F23.T)
 def trans_output(r):
     return np.dot(np.dot(At_F23,r),At_F23.T)

do conv_winof23, conv_direct

 def wino_f23(kernel,input):
     tran_inp = trans_input(input)
     tran_ker = trans_kernel(kernel)
     mid = tran_inp * tran_ker
     out = trans_output(mid)
     return out

 def conv_direct(kernel,input):