megengine.functional.nn.local_response_norm#

local_response_norm(inp, kernel_size=5, k=2.0, alpha=1e-4, beta=0.75)[源代码]#

对输入 Tensor 进行局部响应归一化。

参数:

  • kernel_size (int) – 归一化时核的尺寸。

  • k (float) – 超参数 k。默认：2.0

  • alpha (float) – 超参数 alpha。默认：1e-4

  • beta (float) – 超参数 beta。默认：0.75

示例

>>> import numpy as np
>>> inp = Tensor(np.arange(25, dtype=np.float32).reshape(1,1,5,5))
>>> GT = np.array([[[[ 0.,         0.999925,   1.9994003,  2.9979765,  3.9952066],
...                  [ 4.9906454,  5.983851,   6.974385,   7.961814,   8.945709 ],
...                  [ 9.925651,  10.90122,   11.872011,  12.837625,  13.7976675],
...                  [14.751757,  15.699524,  16.640602,  17.574642,  18.501305 ],
...                  [19.420258,  20.331186,  21.233786,  22.127764,  23.012836 ]]]])
>>> out = F.local_response_norm(inp, kernel_size=3, k=1.0, alpha=1e-4, beta=0.75)
>>> np.testing.assert_allclose(GT, out.numpy(), rtol=1e-6, atol=1e-6)
返回类型:

Tensor