pytorch模型

模型基本定义方法

pytorch中有提供nn.Sequential()、nn.ModuleList()以及nn.ModuleDict()用于集成多个Module，完成模型搭建。其异同如下：

Sequential()	ModuleList() /ModuleDict()
直接搭建网络，定义顺序即为模型连接顺序	List/Dict中元素顺序并不代表其在网络中的真实位置顺序，需要forward函数指定各个层的连接顺序
模型中间无法加入外部输入	模型中间需要之前层的信息的时候，比如 ResNets 中的残差计算，比较方便

通过nn.Sequential()

# 方法一：
import torch.nn as nn
net = nn.Sequential(
        nn.Linear(784, 256),
        nn.ReLU(),
        nn.Linear(256, 10), 
        )
# 方法二：
import collections
net2 = nn.Sequential(collections.OrderedDict([
          ('fc1', nn.Linear(784, 256)),
          ('relu1', nn.ReLU()),
          ('fc2', nn.Linear(256, 10))
          ]))

通过nn.ModuleList()/nn.ModuleDict()

# List
class model(nn.Module):
  def __init__(self):
    super().__init__()
    self.modulelist = nn.ModuleList([nn.Linear(784, 256), nn.ReLU(),nn.Linear(256, 10)])
    
  def forward(self, x):
    for layer in self.modulelist:
      x = layer(x)
    return x
# Dict
class model(nn.Module):
  def __init__(self):
    super().__init__()
    self.moduledict = nn.ModuleDict({
    
    'linear': nn.Linear(784, 256),
    'act': nn.ReLU(),
    'output':nn.Linear(256, 10)
    })
    
  def forward(self, x):
    for layer in self.moduledict:
      x = layer(x)
    return x

复杂模型搭建方法

对于大型复杂模型，可以先将模型分块，然后在进行模型搭建。以U-Net模型为例。

上图为U-Net网络结构，可以分为以下四个模块：

• 每个子块内部的两次卷积（Double Convolution）
• 左侧模型块之间的下采样连接，即最大池化（Max pooling）
• 右侧模型块之间的上采样连接（Up sampling）
• 输出层的处理

模块构建

import torch
import torch.nn as nn
import torch.nn.functional as F

class DoubleConv(nn.Module):
    """(convolution => [BN] => ReLU) * 2"""

    def __init__(self, in_channels, out_channels, mid_channels=None):
        super().__init__()
        if not mid_channels:
            mid_channels = out_channels
        self.double_conv = nn.Sequential(
            nn.Conv2d(in_channels, mid_channels, kernel_size=3, padding=1, bias=False),
            nn.BatchNorm2d(mid_channels),
            nn.ReLU(inplace=True),
            nn.Conv2d(mid_channels, out_channels, kernel_size=3, padding=1, bias=False),
            nn.BatchNorm2d(out_channels),
            nn.ReLU(inplace=True)
        )

    def forward(self, x):
        return self.double_conv(x)

class Down(nn.Module):
    """Downscaling with maxpool then double conv"""

    def __init__(self, in_channels, out_channels):
        super().__init__()
        self.maxpool_conv = nn.Sequential(
            nn.MaxPool2d(2),
            DoubleConv(in_channels, out_channels)
        )

    def forward(self, x):
        return self.maxpool_conv(x)

class Up(nn.Module):
    """Upscaling then double conv"""

    def __init__(self, in_channels, out_channels, bilinear=True):
        super().__init__()

        # if bilinear, use the normal convolutions to reduce the number of channels
        if bilinear:
            self.up = nn.Upsample(scale_factor=2, mode='bilinear', align_corners=True)
            self.conv = DoubleConv(in_channels, out_channels, in_channels // 2)
        else:
            self.up = nn.ConvTranspose2d(in_channels, in_channels // 2, kernel_size=2, stride=2)
            self.conv = DoubleConv(in_channels, out_channels)

    def forward(self, x1, x2):
        x1 = self.up(x1)
        # input is CHW
        diffY = x2.size()[2] - x1.size()[2]
        diffX = x2.size()[3] - x1.size()[3]

        x1 = F.pad(x1, [diffX // 2, diffX - diffX // 2,
                        diffY // 2, diffY - diffY // 2])
        x = torch.cat([x2, x1], dim=1)
        return self.conv(x)

class OutConv(nn.Module):
    def __init__(self, in_channels, out_channels):
        super(OutConv, self).__init__()
        self.conv = nn.Conv2d(in_channels, out_channels, kernel_size=1)

    def forward(self, x):
        return self.conv(x)

模型组装

class UNet(nn.Module):
    def __init__(self, n_channels, n_classes, bilinear=True):
        super(UNet, self).__init__()
        self.n_channels = n_channels
        self.n_classes = n_classes
        self.bilinear = bilinear

        self.inc = DoubleConv(n_channels, 64)
        self.down1 = Down(64, 128)
        self.down2 = Down(128, 256)
        self.down3 = Down(256, 512)
        factor = 2 if bilinear else 1
        self.down4 = Down(512, 1024 // factor)
        self.up1 = Up(1024, 512 // factor, bilinear)
        self.up2 = Up(512, 256 // factor, bilinear)
        self.up3 = Up(256, 128 // factor, bilinear)
        self.up4 = Up(128, 64, bilinear)
        self.outc = OutConv(64, n_classes)

    def forward(self, x):
        x1 = self.inc(x)
        x2 = self.down1(x1)
        x3 = self.down2(x2)
        x4 = self.down3(x3)
        x5 = self.down4(x4)
        x = self.up1(x5, x4)
        x = self.up2(x, x3)
        x = self.up3(x, x2)
        x = self.up4(x, x1)
        logits = self.outc(x)
        return logits

既有模型修改

替换某layer

import torchvision.models as models
net = models.resnet50()
print(net)
# 替换其中fc层
from collections import OrderedDict
classifier = nn.Sequential(OrderedDict([('fc1', nn.Linear(2048, 128)),
                          ('relu1', nn.ReLU()), 
                          ('dropout1',nn.Dropout(0.5)),
                          ('fc2', nn.Linear(128, 10)),
                          ('output', nn.Softmax(dim=1))
                          ]))
#此句直接用定义的classifier替换原来fc层 
net.fc = classifier 

增加输入变量

#定义模型修改
class Model(nn.Module):
    def __init__(self, net):
        super(Model, self).__init__()
        # 原网络结构
        self.net = net
        # 先将2048维的tensor通过激活函数层
        self.relu = nn.ReLU()
        # dropout层
        self.dropout = nn.Dropout(0.5)
        # 全连接层映射到指定的输出维度10
        self.fc_add = nn.Linear(1001, 10, bias=True)
        self.output = nn.Softmax(dim=1)
        
    def forward(self, x, add_variable):
        x = self.net(x)
        #在激活层、dropout层后与外部输入变量拼接
        x = torch.cat((self.dropout(self.relu(x)), add_variable.unsqueeze(1)),1) #unsqueeze操作是为了和net输出的tensor保持维度一致，常用于add_variable是单一数值 (scalar) 的情况
        x = self.fc_add(x)
        x = self.output(x)
        return x
#实例化
model = Model(net).cuda()
#训练
outputs = model(inputs, add_var)

增加输出变量

class Model(nn.Module):
    def __init__(self, net):
        super(Model, self).__init__()
        self.net = net
        self.relu = nn.ReLU()
        self.dropout = nn.Dropout(0.5)
        self.fc1 = nn.Linear(1000, 10, bias=True)
        self.output = nn.Softmax(dim=1)
        
    def forward(self, x, add_variable):
        x1000 = self.net(x)
        x10 = self.dropout(self.relu(x1000))
        x10 = self.fc1(x10)
        x10 = self.output(x10)
        return x10, x1000 #增加输出
model = Model(net).cuda()
out10, out1000 = model(inputs, add_var)

模型保存、加载

PyTorch存储模型主要采用pkl，pt，pth三种格式。
PyTorch模型主要包含两个部分：模型结构和权重。

• 模型：nn.Module的类
• 权重：字典（key是层名，value是权重向量）。

存储也可分为两种形式：

• 存储模型结构+权重
• 只存储权重

from torchvision import models
model = models.resnet152(pretrained=True)

# 保存整个模型
torch.save(model, save_dir)
# 保存模型权重
torch.save(model.state_dict, save_dir)

当出现多GPU并行时存储读取会有单卡、多卡情况，而多卡存储过程名称比单卡多module字段，故当多卡存储时，模型加载会复杂一些。

保存

单卡保存

import os
import torch
from torchvision import models

os.environ['CUDA_VISIBLE_DEVICES'] = '0'   #这里替换成希望使用的GPU编号
model = models.resnet152(pretrained=True)
model.cuda()

# 保存整个模型
torch.save(model, save_dir)

# 保存模型权重
torch.save(model.state_dict(), save_dir)

多卡保存

用nn.DataParallel函数进行分布式训练设置即可

import os
import torch
from torchvision import models

os.environ['CUDA_VISIBLE_DEVICES'] = '0,1,2'   #这里替换成希望使用的GPU编号

model = models.resnet152(pretrained=True)
model = nn.DataParallel(model).cuda()

# 保存整个模型
torch.save(model, save_dir)
# 保存模型权重
torch.save(model.state_dict(), save_dir)

加载

单卡加载

• 单卡保存模型

import os
import torch
from torchvision import models

os.environ['CUDA_VISIBLE_DEVICES'] = '0'   #这里替换成希望使用的GPU编号

# 读取整个模型
loaded_model = torch.load(save_dir)
loaded_model.cuda()

# 读取模型权重
loaded_dict = torch.load(save_dir)
loaded_model = models.resnet152()   #注意这里需要对模型结构有定义
loaded_model.state_dict = loaded_dict
loaded_model.cuda()

• 多卡保存模型

import os
import torch
from torchvision import models

os.environ['CUDA_VISIBLE_DEVICES'] = '0'   #这里替换成希望使用的GPU编号

# 读取整个模型
loaded_model = torch.load(save_dir)
loaded_model = loaded_model.module #不同之处

# 读取模型权重（推荐）
loaded_dict = torch.load(save_dir)
loaded_model = models.resnet152()   #注意这里需要对模型结构有定义
loaded_model = nn.DataParallel(loaded_model).cuda() #不同之处
loaded_model.state_dict = loaded_dict

# 读取模型权重（其他方法1）
from collections import OrderedDict

loaded_dict = torch.load(save_dir)
# 去除module字段
new_state_dict = OrderedDict()
for k, v in loaded_dict.items():
    name = k[7:] # module字段在最前面，从第7个字符开始就可以去掉module
    new_state_dict[name] = v #新字典的key值对应的value一一对应
# 其他与单卡保存模型一致
loaded_model = models.resnet152()  
loaded_model.state_dict = new_state_dict
loaded_model = loaded_model.cuda()

# 读取模型权重（其他方法2）
loaded_model = models.resnet152()    
loaded_dict = torch.load(save_dir)
loaded_model.load_state_dict({
    k.replace('module.', ''): v for k, v in loaded_dict.items()})
loaded_model = loaded_model.cuda()

多卡加载

• 单卡存储模型
  用nn.DataParallel函数进行分布式训练设置即可

import os
import torch
from torchvision import models

os.environ['CUDA_VISIBLE_DEVICES'] = '1,2'   #这里替换成希望使用的GPU编号

# 读取整个模型
loaded_model = torch.load(save_dir)
loaded_model = nn.DataParallel(loaded_model).cuda()#不同处

# 读取模型权重
loaded_dict = torch.load(save_dir)
loaded_model = models.resnet152()   #注意这里需要对模型结构有定义
loaded_model.state_dict = loaded_dict
loaded_model = nn.DataParallel(loaded_model).cuda()#不同处

• 多卡存储模型
  建议仅存储权重，与单卡无异。若只有整个模型，则需要如下代码：

import os
import torch
from torchvision import models

os.environ['CUDA_VISIBLE_DEVICES'] = '0,1,2'   #这里替换成希望使用的GPU编号

loaded_whole_model = torch.load(save_dir)
loaded_model = models.resnet152()   #注意这里需要对模型结构有定义
loaded_model.state_dict = loaded_whole_model.state_dict
loaded_model = nn.DataParallel(loaded_model).cuda()

参考

datawhale：深入浅出pytorch