前言:
博主最近在搭建网络的时候,先搞清楚到底哪些参数进行了训练,网络数据流是不是和自己的预期一样。
更新日志:2022.09.16 1更
一、简单的实例-普通的nn.Sequential可视化
import math
import torch.nn as nn
import torch
from torch.nn import functional as F
from torchsummary import summary
from torchvision import models
def BasicBlock(in_ch,out_ch,stride=1):
return nn.Sequential(
nn.Conv2d(in_ch, out_ch, 3, stride, padding=1, bias=False),
nn.BatchNorm2d(out_ch),
nn.ReLU(inplace=True), # inplace = True原地操作,节省显存
nn.Conv2d(out_ch, out_ch, 3, stride=1, padding=1, bias=False),
nn.BatchNorm2d(out_ch),
nn.ReLU(inplace=True),
)
if __name__ == "__main__":
import imageio
import numpy as np
from torchviz import make_dot
from torch.utils.tensorboard import SummaryWriter
im_data_tensor = torch.random([3,512,512])
model_t = BasicBlock(in_ch=3, out_ch=32)
out = model_t(im_data_tensor)
# 1. 来用tensorflow进行可视化
with SummaryWriter("./log", comment="sample_model_visualization") as sw:
sw.add_graph(model_t, im_data_tensor)
应该不难看出其中的逻辑关系,每一个操作后面都有一个标号,
二、网络的结构和forward息息相关,
并行
def forward(self, x):
x0 = self.conv_br1(x)
x1 = self.conv_br2(x)
return x0,x1
串行
def forward(self, x):
x0 = self.conv_br1(x)
x1 = self.conv_br2(x0)
return x1
与最终的输出无关的变量或者计算流,不会出现在网络中
def forward(self, x):
x0 = self.conv_br1(x)
x1 = self.conv_br2(x)
return x1