from pathlib import Path
from typing import Optionalimport torch
import torch.nn as nn
from torch import Tensorclass BN(nn.Module):def __init__(self,num_features,momentum=0.1,eps=1e-8):##num_features是通道数"""初始化方法:param num_features:特征属性的数量，也就是通道数目C"""super(BN, self).__init__()##register_buffer：将属性当成parameter进行处理，唯一的区别就是不参与反向传播的梯度求解self.register_buffer('running_mean', torch.zeros(1, num_features, 1, 1))self.register_buffer('running_var', torch.zeros(1, num_features, 1, 1))self.running_mean: Optional[Tensor]self.running_var: Optional[Tensor]self.running_mean=torch.zeros([1,num_features,1,1])self.running_var=torch.zeros([1,num_features,1,1])self.gamma=nn.Parameter(torch.ones([1,num_features,1,1]))self.beta=nn.Parameter(torch.zeros(1,num_features,1,1))self.eps=epsself.momentum=momentumdef forward(self,x):"""前向过程output=(x-μ）/α*γ+β:param x: [N,C,H,W]:return: [N,C,H,W]"""if self.training:#训练阶段--》使用当前批次的数据_mean=torch.mean(x,dim=(0,2,3),keepdim=True)_var = torch.var(x, dim=(0, 2, 3), keepdim=True)#将训练过程中的均值和方差保存下来--方便推理的时候使用--》滑动平均self.running_mean=self.momentum*self.running_mean+(1.0-self.momentum)*_meanself.running_var=self.momentum*self.running_var+(1.0-self.momentum)*_varelse:#推理阶段-->使用的是训练过程中的累积数据_mean=self.running_mean_var=self.running_varz=(x-_mean)/torch.sqrt(_var+self.eps)*self.gamma+self.betareturn zif __name__ == '__main__':torch.manual_seed(28)path_dir=Path("./output/models")path_dir.mkdir(parents=True,exist_ok=True)device=torch.device("cuda" if torch.cuda.is_available() else "cpu")bn=BN(num_features=12)bn.to(device)#只针对子模块和参数进行转换#模拟训练过程bn.train()xs=[torch.randn(8,12,32,32).to(device) for _ in range(10)]for _x in xs:bn(_x)print(bn.running_mean.view(-1))print(bn.running_var.view(-1))#模拟推理过程bn.eval()_r=bn(xs[0])print(_r.shape)bn=bn.cpu()#保存都是以cpu保存,恢复再自己转回GPU上#模拟模型保存torch.save(bn,str(path_dir/'bn_model.pkl'))#state_dict：获取当前模块的所有参数(Parameter+register_buffer)torch.save(bn.state_dict(),str(path_dir/"bn_params.pkl"))#pt结构的保存traced_script_module=torch.jit.trace(bn.eval(),xs[0].cpu())traced_script_module.save("./output/bn_model.pt")#模拟模型恢复bn_model=torch.load(str(path_dir/"bn_model.pkl"),map_location='cpu')bn_params=torch.load(str(path_dir/"bn_params.pkl"),map_location='cpu')print(len(bn_params))
