PyTorch实现ResNet50、ResNet101和ResNet152示例

时间:2022-09-16 12:13:30

PyTorch: https://github.com/shanglianlm0525/PyTorch-Networks

PyTorch实现ResNet50、ResNet101和ResNet152示例

?
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
import torch
import torch.nn as nn
import torchvision
import numpy as np
 
print("PyTorch Version: ",torch.__version__)
print("Torchvision Version: ",torchvision.__version__)
 
__all__ = ['ResNet50', 'ResNet101','ResNet152']
 
def Conv1(in_planes, places, stride=2):
  return nn.Sequential(
    nn.Conv2d(in_channels=in_planes,out_channels=places,kernel_size=7,stride=stride,padding=3, bias=False),
    nn.BatchNorm2d(places),
    nn.ReLU(inplace=True),
    nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
  )
 
class Bottleneck(nn.Module):
  def __init__(self,in_places,places, stride=1,downsampling=False, expansion = 4):
    super(Bottleneck,self).__init__()
    self.expansion = expansion
    self.downsampling = downsampling
 
    self.bottleneck = nn.Sequential(
      nn.Conv2d(in_channels=in_places,out_channels=places,kernel_size=1,stride=1, bias=False),
      nn.BatchNorm2d(places),
      nn.ReLU(inplace=True),
      nn.Conv2d(in_channels=places, out_channels=places, kernel_size=3, stride=stride, padding=1, bias=False),
      nn.BatchNorm2d(places),
      nn.ReLU(inplace=True),
      nn.Conv2d(in_channels=places, out_channels=places*self.expansion, kernel_size=1, stride=1, bias=False),
      nn.BatchNorm2d(places*self.expansion),
    )
 
    if self.downsampling:
      self.downsample = nn.Sequential(
        nn.Conv2d(in_channels=in_places, out_channels=places*self.expansion, kernel_size=1, stride=stride, bias=False),
        nn.BatchNorm2d(places*self.expansion)
      )
    self.relu = nn.ReLU(inplace=True)
  def forward(self, x):
    residual = x
    out = self.bottleneck(x)
 
    if self.downsampling:
      residual = self.downsample(x)
 
    out += residual
    out = self.relu(out)
    return out
 
class ResNet(nn.Module):
  def __init__(self,blocks, num_classes=1000, expansion = 4):
    super(ResNet,self).__init__()
    self.expansion = expansion
 
    self.conv1 = Conv1(in_planes = 3, places= 64)
 
    self.layer1 = self.make_layer(in_places = 64, places= 64, block=blocks[0], stride=1)
    self.layer2 = self.make_layer(in_places = 256,places=128, block=blocks[1], stride=2)
    self.layer3 = self.make_layer(in_places=512,places=256, block=blocks[2], stride=2)
    self.layer4 = self.make_layer(in_places=1024,places=512, block=blocks[3], stride=2)
 
    self.avgpool = nn.AvgPool2d(7, stride=1)
    self.fc = nn.Linear(2048,num_classes)
 
    for m in self.modules():
      if isinstance(m, nn.Conv2d):
        nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
      elif isinstance(m, nn.BatchNorm2d):
        nn.init.constant_(m.weight, 1)
        nn.init.constant_(m.bias, 0)
 
  def make_layer(self, in_places, places, block, stride):
    layers = []
    layers.append(Bottleneck(in_places, places,stride, downsampling =True))
    for i in range(1, block):
      layers.append(Bottleneck(places*self.expansion, places))
 
    return nn.Sequential(*layers)
 
 
  def forward(self, x):
    x = self.conv1(x)
 
    x = self.layer1(x)
    x = self.layer2(x)
    x = self.layer3(x)
    x = self.layer4(x)
 
    x = self.avgpool(x)
    x = x.view(x.size(0), -1)
    x = self.fc(x)
    return x
 
def ResNet50():
  return ResNet([3, 4, 6, 3])
 
def ResNet101():
  return ResNet([3, 4, 23, 3])
 
def ResNet152():
  return ResNet([3, 8, 36, 3])
 
 
if __name__=='__main__':
  #model = torchvision.models.resnet50()
  model = ResNet50()
  print(model)
 
  input = torch.randn(1, 3, 224, 224)
  out = model(input)
  print(out.shape)

以上这篇PyTorch实现ResNet50、ResNet101和ResNet152示例就是小编分享给大家的全部内容了,希望能给大家一个参考,也希望大家多多支持服务器之家。

原文链接:https://blog.csdn.net/shanglianlm/article/details/86376627