nn.BatchNorm2d的具体实现

参考：https://blog.csdn.net/qq_38253797/article/details/116847588

 

import torch
import torch.nn as nn
import torch.nn.functional as F
import numpy as np

def _bn():
    _batch = torch.randn(3, 4, 5, 5)
    aa = []
    bb = []
    for c in range(4):
        aa.append(0 + torch.mean(_batch[:, c, :, :]) * 0.1)
        bb.append(1 * 0.9 + torch.var(_batch[:, c, :, :]) * 0.1)
    print(aa)
    print(bb)

    m = nn.BatchNorm2d(4, affine=False, momentum=0.1)
    _a1 = m(_batch)
    print(_a1.shape)
    print(m.running_mean)
    print(m.running_var)


if __name__ == '__main__':
    _bn()