脚本之家，脚本语言编程技术及教程分享平台！
分类导航

Python|VBS|Ruby|Lua|perl|VBA|Golang|PowerShell|Erlang|autoit|Dos|bat|

服务器之家 - 脚本之家 - Python - pytorch构建多模型实例

pytorch构建多模型实例

2020-04-22 09:52朴素.无恙 Python

今天小编就为大家分享一篇pytorch构建多模型实例，具有很好的参考价值，希望对大家有所帮助。一起跟随小编过来看看吧

pytorch构建双模型

第一部分：构建"se_resnet152","DPN92()"双模型

?
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
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
import numpy as np
from functools import partial
import torch
from torch import nn
import torch.nn.functional as F
from torch.optim import SGD,Adam
from torch.autograd import Variable
from torch.utils.data import Dataset, DataLoader
 
from torch.optim.optimizer import Optimizer
 
import torchvision
from torchvision import models
import pretrainedmodels
from pretrainedmodels.models import *
from torch import nn
from torchvision import transforms as T
import random
 
 
 
random.seed(2050)
np.random.seed(2050)
torch.manual_seed(2050)
torch.cuda.manual_seed_all(2050)
 
class FCViewer(nn.Module):
  def forward(self, x):
    return x.view(x.size(0), -1)
 
  
'''Dual Path Networks in PyTorch.'''
class Bottleneck(nn.Module):
  def __init__(self, last_planes, in_planes, out_planes, dense_depth, stride, first_layer):
    super(Bottleneck, self).__init__()
    self.out_planes = out_planes
    self.dense_depth = dense_depth
 
    self.conv1 = nn.Conv2d(last_planes, in_planes, kernel_size=1, bias=False)
    self.bn1 = nn.BatchNorm2d(in_planes)
    self.conv2 = nn.Conv2d(in_planes, in_planes, kernel_size=3, stride=stride, padding=1, groups=32, bias=False)
    self.bn2 = nn.BatchNorm2d(in_planes)
    self.conv3 = nn.Conv2d(in_planes, out_planes+dense_depth, kernel_size=1, bias=False)
    self.bn3 = nn.BatchNorm2d(out_planes+dense_depth)
 
    self.shortcut = nn.Sequential()
    if first_layer:
      self.shortcut = nn.Sequential(
        nn.Conv2d(last_planes, out_planes+dense_depth, kernel_size=1, stride=stride, bias=False),
        nn.BatchNorm2d(out_planes+dense_depth)
      )
 
  def forward(self, x):
    out = F.relu(self.bn1(self.conv1(x)))
    out = F.relu(self.bn2(self.conv2(out)))
    out = self.bn3(self.conv3(out))
    x = self.shortcut(x)
    d = self.out_planes
    out = torch.cat([x[:,:d,:,:]+out[:,:d,:,:], x[:,d:,:,:], out[:,d:,:,:]], 1)
    out = F.relu(out)
    return out
 
 
class DPN(nn.Module):
  def __init__(self, cfg):
    super(DPN, self).__init__()
    in_planes, out_planes = cfg['in_planes'], cfg['out_planes']
    num_blocks, dense_depth = cfg['num_blocks'], cfg['dense_depth']
 
    self.conv1 = nn.Conv2d(7, 64, kernel_size=3, stride=1, padding=1, bias=False)
    self.bn1 = nn.BatchNorm2d(64)
    self.last_planes = 64
    self.layer1 = self._make_layer(in_planes[0], out_planes[0], num_blocks[0], dense_depth[0], stride=1)
    self.layer2 = self._make_layer(in_planes[1], out_planes[1], num_blocks[1], dense_depth[1], stride=2)
    self.layer3 = self._make_layer(in_planes[2], out_planes[2], num_blocks[2], dense_depth[2], stride=2)
    self.layer4 = self._make_layer(in_planes[3], out_planes[3], num_blocks[3], dense_depth[3], stride=2)
    self.linear = nn.Linear(out_planes[3]+(num_blocks[3]+1)*dense_depth[3], 64)
    self.bn2 = nn.BatchNorm1d(64)
  def _make_layer(self, in_planes, out_planes, num_blocks, dense_depth, stride):
    strides = [stride] + [1]*(num_blocks-1)
    layers = []
    for i,stride in enumerate(strides):
      layers.append(Bottleneck(self.last_planes, in_planes, out_planes, dense_depth, stride, i==0))
      self.last_planes = out_planes + (i+2) * dense_depth
    return nn.Sequential(*layers)
 
  def forward(self, x):
    out = F.relu(self.bn1(self.conv1(x)))
    out = self.layer1(out)
    out = self.layer2(out)
    out = self.layer3(out)
    out = self.layer4(out)
    out = F.avg_pool2d(out, 4)
    out = out.view(out.size(0), -1)
    out = self.linear(out)
    out= F.relu(self.bn2(out))
    return out
 
 
 
def DPN26():
  cfg = {
    'in_planes': (96,192,384,768),
    'out_planes': (256,512,1024,2048),
    'num_blocks': (2,2,2,2),
    'dense_depth': (16,32,24,128)
  }
  return DPN(cfg)
 
def DPN92():
  cfg = {
    'in_planes': (96,192,384,768),
    'out_planes': (256,512,1024,2048),
    'num_blocks': (3,4,20,3),
    'dense_depth': (16,32,24,128)
  }
  return DPN(cfg)
class MultiModalNet(nn.Module):
  def __init__(self, backbone1, backbone2, drop, pretrained=True):
    super().__init__()
    if pretrained:
      img_model = pretrainedmodels.__dict__[backbone1](num_classes=1000, pretrained='imagenet') #seresnext101
    else:
      img_model = pretrainedmodels.__dict__[backbone1](num_classes=1000, pretrained=None)
    
    self.visit_model=DPN26()
    
    self.img_encoder = list(img_model.children())[:-2]
    self.img_encoder.append(nn.AdaptiveAvgPool2d(1))
    
    self.img_encoder = nn.Sequential(*self.img_encoder)
    if drop > 0:
      self.img_fc = nn.Sequential(FCViewer(),
                  nn.Dropout(drop),
                  nn.Linear(img_model.last_linear.in_features, 512),
                  nn.BatchNorm1d(512))
                  
    else:
      self.img_fc = nn.Sequential(
        FCViewer(),
        nn.BatchNorm1d(img_model.last_linear.in_features),
        nn.Linear(img_model.last_linear.in_features, 512))
    self.bn=nn.BatchNorm1d(576)
    self.cls = nn.Linear(576,9)
 
  def forward(self, x_img,x_vis):
    x_img = self.img_encoder(x_img)
    x_img = self.img_fc(x_img)
    x_vis=self.visit_model(x_vis)
    x_cat = torch.cat((x_img,x_vis),1)
    x_cat = F.relu(self.bn(x_cat))
    x_cat = self.cls(x_cat)
    return x_cat
 
test_x = Variable(torch.zeros(64, 7,26,24))
test_x1 = Variable(torch.zeros(64, 3,224,224))
model=MultiModalNet("se_resnet152","DPN92()",0.1)
out=model(test_x1,test_x)
print(model._modules.keys())
print(model)
 
print(out.shape)

第二部分构建densenet201单模型

?
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
#encoding:utf-8
import torchvision.models as models
import torch
import pretrainedmodels
from torch import nn
from torch.autograd import Variable
#model = models.resnet18(pretrained=True)
#print(model)
#print(model._modules.keys())
#feature = torch.nn.Sequential(*list(model.children())[:-2])#模型的结构
#print(feature)
'''
class FCViewer(nn.Module):
  def forward(self, x):
    return x.view(x.size(0), -1)
class M(nn.Module):
  def __init__(self, backbone1, drop, pretrained=True):
    super(M,self).__init__()
    if pretrained:
      img_model = pretrainedmodels.__dict__[backbone1](num_classes=1000, pretrained='imagenet')
    else:
      img_model = pretrainedmodels.__dict__[backbone1](num_classes=1000, pretrained=None)
    
    self.img_encoder = list(img_model.children())[:-1]
    self.img_encoder.append(nn.AdaptiveAvgPool2d(1))
    self.img_encoder = nn.Sequential(*self.img_encoder)
 
    if drop > 0:
      self.img_fc = nn.Sequential(FCViewer(),
                  nn.Dropout(drop),
                  nn.Linear(img_model.last_linear.in_features, 236))
                  
    else:
      self.img_fc = nn.Sequential(
        FCViewer(),
        nn.Linear(img_model.last_linear.in_features, 236)
      )
 
    self.cls = nn.Linear(236,9)
 
  def forward(self, x_img):
    x_img = self.img_encoder(x_img)
    x_img = self.img_fc(x_img)
    return x_img
 
model1=M('densenet201',0,pretrained=True)
print(model1)
print(model1._modules.keys())
feature = torch.nn.Sequential(*list(model1.children())[:-2])#模型的结构
feature1 = torch.nn.Sequential(*list(model1.children())[:])
#print(feature)
#print(feature1)
test_x = Variable(torch.zeros(1, 3, 100, 100))
out=feature(test_x)
print(out.shape)
'''
'''
import torch.nn.functional as F
class LenetNet(nn.Module):
  def __init__(self):
    super(LenetNet, self).__init__()
    self.conv1 = nn.Conv2d(7, 6, 5)
    self.conv2 = nn.Conv2d(6, 16, 5)
    self.fc1  = nn.Linear(144, 120)
    self.fc2  = nn.Linear(120, 84)
    self.fc3  = nn.Linear(84, 10)
  def forward(self, x):
    x = F.max_pool2d(F.relu(self.conv1(x)), (2, 2))
    x = F.max_pool2d(F.relu(self.conv2(x)), 2)
    x = x.view(x.size()[0], -1)
    x = F.relu(self.fc1(x))
    x = F.relu(self.fc2(x))
    x = self.fc3(x)   
    return x
 
model1=LenetNet()
#print(model1)
#print(model1._modules.keys())
feature = torch.nn.Sequential(*list(model1.children())[:-3])#模型的结构
#feature1 = torch.nn.Sequential(*list(model1.children())[:])
print(feature)
#print(feature1)
test_x = Variable(torch.zeros(1, 7, 27, 24))
out=model1(test_x)
print(out.shape)
 
class FCViewer(nn.Module):
  def forward(self, x):
    return x.view(x.size(0), -1)
class M(nn.Module):
  def __init__(self):
    super(M,self).__init__()
    img_model =model1
    self.img_encoder = list(img_model.children())[:-3]
    self.img_encoder.append(nn.AdaptiveAvgPool2d(1))
    self.img_encoder = nn.Sequential(*self.img_encoder)
    self.img_fc = nn.Sequential(FCViewer(),
              nn.Linear(16, 236))
    self.cls = nn.Linear(236,9)
 
  def forward(self, x_img):
    x_img = self.img_encoder(x_img)
    x_img = self.img_fc(x_img)
    return x_img
 
model2=M()
 
test_x = Variable(torch.zeros(1, 7, 27, 24))
out=model2(test_x)
print(out.shape)
 
'''

以上这篇pytorch构建多模型实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/weixin_40123108/article/details/90670584

  • Pytorch
  • 多模型

    • 延伸 · 阅读

    精彩推荐
    最近更新
    编辑推荐
    阅读排行
    热门标签
    2048 　 6006 　 contextlib 　 SLOTS 　 MELIAE 　 FEEDPARSER 　 暂存 　 持久性管理 　 pickle模块 　 THREADING 　 reduce 　 nignx 　 装饰器 　 函数式编程 　 兔子毒药 　 监控文件 　 流量监控 　 Web框架 　 办公工具 　 PIL库 　 高斯模糊 　 wxPython 　 import 　 自定义模块 　 Python开发 　 元类 　 metaclass 　 魔术方法 　 DNSPod 　 动态解析域名 　
    © 2019-2023 脚本之家 | 服务器之家(www.zzvips.com)旗下站点 版权所有关于我们联系我们版权申明网站地图