模型部分最近在做一个特征融合，需要将卷积网络的最后一层的深度特征拼接起来(concat)，搭建网络的时候遇到了一系列问题，记录方便以后查看
from torch.nn import functional as F
import torch.nn as nn
import torch
import torchvision
class MyEnsemble(nn.Module):
def __init__(self, nb_classes=2):
super(MyEnsemble, self).__init__()
self.model1 = torchvision.models.resnet18(pretrained=False)
self.model2 = torchvision.models.resnet18(pretrained=False)
self.model3 = torchvision.models.resnet18(pretrained=False)
# Freeze the models
for param in self.model1.parameters():
param.requires_grad = True
for param in self.model2.parameters():
param.requires_grad = True
for param in self.model3.parameters():
param.requires_grad = True
# create a new classifier
self.model1.fc = nn.Identity()
self.model2.fc = nn.Identity()
self.model3.fc = nn.Identity()
self.classifier = nn.Linear(512*3, nb_classes)
def forward(self, x1, x2, x3):
x1 = self.model1(x1),
print(x1)
x1 = x1.view(x1.size(0), -1),
x2 = self.model2(x2),
x2 = x2.view(x2.size(0), -1),
x3 = self.model3(x3),
x3 = x3.view(x3.size(0), -1),
x = torch.cat((x1, x2, x3), dim=1),
x = self.classifier(F.relu(x))
return x
x1 = torch.randn(2, 3, 224, 224)
x2 = torch.randn(2, 3, 224, 224)
x3 = torch.randn(2, 3, 224, 224)
model = MyEnsemble()
outputs = model(x1, x2, x3)
print(outputs)
然后在测试输出时，得到了报错：AttributeError: 'tuple' object has no attribute 'view'，报错的原因是网络经过pooling层之后的输出是一个tuple类型：如下为输出结果（print大法是真的香）
此时我们要做的是，将tuple里面的tensor提取出来，由于这里tuple里面只有一个元素，一个简单的语法就可以了：x1=x1[0]，以下为将tuple中tensor提取出来后的前后对比
又认真研究了一下，原来是我太菜，在forward方法里面，每个语句后面都加了逗号，是不需要逗号的啊喂，好菜好菜！写在这里警醒自己，不过话说回来tuple提取方法也是可以的
除此之外，还有两个地方最初不是很理解
for param in self.model1.parameters():
param.requires_grad = True
self.model3.fc = nn.Identity()
（1）param.requires_grad = True，表示允许参数更新，也就是说允许你在训练时通过反向传播来更新参数，反之就是冻结
（2）self.model3.fc = nn.Identity()，去pytorch的论坛找了一下，别人给出的解释是 “该模块将只返回输入，无需任何操作，可用于例如替换其他层”，我的理解是，在你需要替换掉网络时，你需要使用该语句nn.Identity()，将原有的网络结构保持，然后替换掉你想替换的层例如我的
self.model1.fc = nn.Identity()
self.model2.fc = nn.Identity()
self.model3.fc = nn.Identity()
self.classifier = nn.Linear(512*3, nb_classes)
就是将卷积操作保持，然后将三个深度特征融合以后用一个新的全连接层替换掉原有全连接层
然后接下来我要查看我的模型到底是的组成的，有没有按照我预想的来：
torchsummary.summary方法
完整代码：
模型部分：
from torch.nn import functional as F
import torch.nn as nn
import torch
import torchvision
class MyEnsemble(nn.Module):
def __init__(self, nb_classes=2):
super(MyEnsemble, self).__init__()
self.model1 = torchvision.models.resnet18(pretrained=False)
self.model2 = torchvision.models.resnet18(pretrained=False)
self.model3 = torchvision.models.resnet18(pretrained=False)
# Freeze the models
for param in self.model1.parameters():
param.requires_grad = True
for param in self.model2.parameters():
param.requires_grad = True
for param in self.model3.parameters():
param.requires_grad = True
# create a new classifier
self.model1.fc = nn.Identity()
self.model2.fc = nn.Identity()
self.model3.fc = nn.Identity()
self.classifier = nn.Linear(512*3, nb_classes)
def forward(self, x1, x2, x3):
x1 = self.model1(x1)
# x1 = x1[0]
x1 = x1.view(x1.size(0), -1)
x2 = self.model2(x2)
# x2 = x2[0]
x2 = x2.view(x2.size(0), -1)
x3 = self.model3(x3)
# x3 = x3[0]
x3 = x3.view(x3.size(0), -1)
x = torch.cat((x1, x2, x3), dim=1)
x = self.classifier(F.relu(x))
return x
查看输出结果以及网络结构
import torch
from my_model import MyEnsemble
from torchsummary import summary
def train():
device = torch.device('cuda:0')
x1 = torch.randn(10, 3, 224, 224)
x2 = torch.randn(10, 3, 224, 224)
x3 = torch.randn(10, 3, 224, 224)
model = MyEnsemble()
outputs = model(x1, x2, x3)
print(outputs)
model.to(device)
summary(model, [(3, 224, 224), (3, 224, 224), (3, 224, 224)])
if __name__=='__main__':
train()
输出结果如下：可以看到完整的网络结构
tensor([[ 0.3332, -0.7907],
[ 0.1526, -0.9038],
[ 0.2006, -0.9861],
[ 0.0889, -0.8982],
[ 0.2047, -0.7573],
[ 0.0432, -0.7982],
[ 0.1041, -0.9219],
[ 0.1470, -0.8201],
[ 0.2360, -0.9237],
[ 0.2444, -0.8210]], grad_fn=)
----------------------------------------------------------------
Layer (type) Output Shape Param #
================================================================
Conv2d-1 [-1, 64, 112, 112] 9,408
BatchNorm2d-2 [-1, 64, 112, 112] 128
ReLU-3 [-1, 64, 112, 112] 0
MaxPool2d-4 [-1, 64, 56, 56] 0
Conv2d-5 [-1, 64, 56, 56] 36,864
BatchNorm2d-6 [-1, 64, 56, 56] 128
ReLU-7 [-1, 64, 56, 56] 0
Conv2d-8 [-1, 64, 56, 56] 36,864
BatchNorm2d-9 [-1, 64, 56, 56] 128
ReLU-10 [-1, 64, 56, 56] 0
BasicBlock-11 [-1, 64, 56, 56] 0
Conv2d-12 [-1, 64, 56, 56] 36,864
BatchNorm2d-13 [-1, 64, 56, 56] 128
ReLU-14 [-1, 64, 56, 56] 0
Conv2d-15 [-1, 64, 56, 56] 36,864
BatchNorm2d-16 [-1, 64, 56, 56] 128
ReLU-17 [-1, 64, 56, 56] 0
BasicBlock-18 [-1, 64, 56, 56] 0
Conv2d-19 [-1, 128, 28, 28] 73,728
BatchNorm2d-20 [-1, 128, 28, 28] 256
ReLU-21 [-1, 128, 28, 28] 0
Conv2d-22 [-1, 128, 28, 28] 147,456
BatchNorm2d-23 [-1, 128, 28, 28] 256
Conv2d-24 [-1, 128, 28, 28] 8,192
BatchNorm2d-25 [-1, 128, 28, 28] 256
ReLU-26 [-1, 128, 28, 28] 0
BasicBlock-27 [-1, 128, 28, 28] 0
Conv2d-28 [-1, 128, 28, 28] 147,456
BatchNorm2d-29 [-1, 128, 28, 28] 256
ReLU-30 [-1, 128, 28, 28] 0
Conv2d-31 [-1, 128, 28, 28] 147,456
BatchNorm2d-32 [-1, 128, 28, 28] 256
ReLU-33 [-1, 128, 28, 28] 0
BasicBlock-34 [-1, 128, 28, 28] 0
Conv2d-35 [-1, 256, 14, 14] 294,912
BatchNorm2d-36 [-1, 256, 14, 14] 512
ReLU-37 [-1, 256, 14, 14] 0
Conv2d-38 [-1, 256, 14, 14] 589,824
BatchNorm2d-39 [-1, 256, 14, 14] 512
Conv2d-40 [-1, 256, 14, 14] 32,768
BatchNorm2d-41 [-1, 256, 14, 14] 512
ReLU-42 [-1, 256, 14, 14] 0
BasicBlock-43 [-1, 256, 14, 14] 0
Conv2d-44 [-1, 256, 14, 14] 589,824
BatchNorm2d-45 [-1, 256, 14, 14] 512
ReLU-46 [-1, 256, 14, 14] 0
Conv2d-47 [-1, 256, 14, 14] 589,824
BatchNorm2d-48 [-1, 256, 14, 14] 512
ReLU-49 [-1, 256, 14, 14] 0
BasicBlock-50 [-1, 256, 14, 14] 0
Conv2d-51 [-1, 512, 7, 7] 1,179,648
BatchNorm2d-52 [-1, 512, 7, 7] 1,024
ReLU-53 [-1, 512, 7, 7] 0
Conv2d-54 [-1, 512, 7, 7] 2,359,296
BatchNorm2d-55 [-1, 512, 7, 7] 1,024
Conv2d-56 [-1, 512, 7, 7] 131,072
BatchNorm2d-57 [-1, 512, 7, 7] 1,024
ReLU-58 [-1, 512, 7, 7] 0
BasicBlock-59 [-1, 512, 7, 7] 0
Conv2d-60 [-1, 512, 7, 7] 2,359,296
BatchNorm2d-61 [-1, 512, 7, 7] 1,024
ReLU-62 [-1, 512, 7, 7] 0
Conv2d-63 [-1, 512, 7, 7] 2,359,296
BatchNorm2d-64 [-1, 512, 7, 7] 1,024
ReLU-65 [-1, 512, 7, 7] 0
BasicBlock-66 [-1, 512, 7, 7] 0
AdaptiveAvgPool2d-67 [-1, 512, 1, 1] 0
Identity-68 [-1, 512] 0
ResNet-69 [-1, 512] 0
Conv2d-70 [-1, 64, 112, 112] 9,408
BatchNorm2d-71 [-1, 64, 112, 112] 128
ReLU-72 [-1, 64, 112, 112] 0
MaxPool2d-73 [-1, 64, 56, 56] 0
Conv2d-74 [-1, 64, 56, 56] 36,864
BatchNorm2d-75 [-1, 64, 56, 56] 128
ReLU-76 [-1, 64, 56, 56] 0
Conv2d-77 [-1, 64, 56, 56] 36,864
BatchNorm2d-78 [-1, 64, 56, 56] 128
ReLU-79 [-1, 64, 56, 56] 0
BasicBlock-80 [-1, 64, 56, 56] 0
Conv2d-81 [-1, 64, 56, 56] 36,864
BatchNorm2d-82 [-1, 64, 56, 56] 128
ReLU-83 [-1, 64, 56, 56] 0
Conv2d-84 [-1, 64, 56, 56] 36,864
BatchNorm2d-85 [-1, 64, 56, 56] 128
ReLU-86 [-1, 64, 56, 56] 0
BasicBlock-87 [-1, 64, 56, 56] 0
Conv2d-88 [-1, 128, 28, 28] 73,728
BatchNorm2d-89 [-1, 128, 28, 28] 256
ReLU-90 [-1, 128, 28, 28] 0
Conv2d-91 [-1, 128, 28, 28] 147,456
BatchNorm2d-92 [-1, 128, 28, 28] 256
Conv2d-93 [-1, 128, 28, 28] 8,192
BatchNorm2d-94 [-1, 128, 28, 28] 256
ReLU-95 [-1, 128, 28, 28] 0
BasicBlock-96 [-1, 128, 28, 28] 0
Conv2d-97 [-1, 128, 28, 28] 147,456
BatchNorm2d-98 [-1, 128, 28, 28] 256
ReLU-99 [-1, 128, 28, 28] 0
Conv2d-100 [-1, 128, 28, 28] 147,456
BatchNorm2d-101 [-1, 128, 28, 28] 256
ReLU-102 [-1, 128, 28, 28] 0
BasicBlock-103 [-1, 128, 28, 28] 0
Conv2d-104 [-1, 256, 14, 14] 294,912
BatchNorm2d-105 [-1, 256, 14, 14] 512
ReLU-106 [-1, 256, 14, 14] 0
Conv2d-107 [-1, 256, 14, 14] 589,824
BatchNorm2d-108 [-1, 256, 14, 14] 512
Conv2d-109 [-1, 256, 14, 14] 32,768
BatchNorm2d-110 [-1, 256, 14, 14] 512
ReLU-111 [-1, 256, 14, 14] 0
BasicBlock-112 [-1, 256, 14, 14] 0
Conv2d-113 [-1, 256, 14, 14] 589,824
BatchNorm2d-114 [-1, 256, 14, 14] 512
ReLU-115 [-1, 256, 14, 14] 0
Conv2d-116 [-1, 256, 14, 14] 589,824
BatchNorm2d-117 [-1, 256, 14, 14] 512
ReLU-118 [-1, 256, 14, 14] 0
BasicBlock-119 [-1, 256, 14, 14] 0
Conv2d-120 [-1, 512, 7, 7] 1,179,648
BatchNorm2d-121 [-1, 512, 7, 7] 1,024
ReLU-122 [-1, 512, 7, 7] 0
Conv2d-123 [-1, 512, 7, 7] 2,359,296
BatchNorm2d-124 [-1, 512, 7, 7] 1,024
Conv2d-125 [-1, 512, 7, 7] 131,072
BatchNorm2d-126 [-1, 512, 7, 7] 1,024
ReLU-127 [-1, 512, 7, 7] 0
BasicBlock-128 [-1, 512, 7, 7] 0
Conv2d-129 [-1, 512, 7, 7] 2,359,296
BatchNorm2d-130 [-1, 512, 7, 7] 1,024
ReLU-131 [-1, 512, 7, 7] 0
Conv2d-132 [-1, 512, 7, 7] 2,359,296
BatchNorm2d-133 [-1, 512, 7, 7] 1,024
ReLU-134 [-1, 512, 7, 7] 0
BasicBlock-135 [-1, 512, 7, 7] 0
AdaptiveAvgPool2d-136 [-1, 512, 1, 1] 0
Identity-137 [-1, 512] 0
ResNet-138 [-1, 512] 0
Conv2d-139 [-1, 64, 112, 112] 9,408
BatchNorm2d-140 [-1, 64, 112, 112] 128
ReLU-141 [-1, 64, 112, 112] 0
MaxPool2d-142 [-1, 64, 56, 56] 0
Conv2d-143 [-1, 64, 56, 56] 36,864
BatchNorm2d-144 [-1, 64, 56, 56] 128
ReLU-145 [-1, 64, 56, 56] 0
Conv2d-146 [-1, 64, 56, 56] 36,864
BatchNorm2d-147 [-1, 64, 56, 56] 128
ReLU-148 [-1, 64, 56, 56] 0
BasicBlock-149 [-1, 64, 56, 56] 0
Conv2d-150 [-1, 64, 56, 56] 36,864
BatchNorm2d-151 [-1, 64, 56, 56] 128
ReLU-152 [-1, 64, 56, 56] 0
Conv2d-153 [-1, 64, 56, 56] 36,864
BatchNorm2d-154 [-1, 64, 56, 56] 128
ReLU-155 [-1, 64, 56, 56] 0
BasicBlock-156 [-1, 64, 56, 56] 0
Conv2d-157 [-1, 128, 28, 28] 73,728
BatchNorm2d-158 [-1, 128, 28, 28] 256
ReLU-159 [-1, 128, 28, 28] 0
Conv2d-160 [-1, 128, 28, 28] 147,456
BatchNorm2d-161 [-1, 128, 28, 28] 256
Conv2d-162 [-1, 128, 28, 28] 8,192
BatchNorm2d-163 [-1, 128, 28, 28] 256
ReLU-164 [-1, 128, 28, 28] 0
BasicBlock-165 [-1, 128, 28, 28] 0
Conv2d-166 [-1, 128, 28, 28] 147,456
BatchNorm2d-167 [-1, 128, 28, 28] 256
ReLU-168 [-1, 128, 28, 28] 0
Conv2d-169 [-1, 128, 28, 28] 147,456
BatchNorm2d-170 [-1, 128, 28, 28] 256
ReLU-171 [-1, 128, 28, 28] 0
BasicBlock-172 [-1, 128, 28, 28] 0
Conv2d-173 [-1, 256, 14, 14] 294,912
BatchNorm2d-174 [-1, 256, 14, 14] 512
ReLU-175 [-1, 256, 14, 14] 0
Conv2d-176 [-1, 256, 14, 14] 589,824
BatchNorm2d-177 [-1, 256, 14, 14] 512
Conv2d-178 [-1, 256, 14, 14] 32,768
BatchNorm2d-179 [-1, 256, 14, 14] 512
ReLU-180 [-1, 256, 14, 14] 0
BasicBlock-181 [-1, 256, 14, 14] 0
Conv2d-182 [-1, 256, 14, 14] 589,824
BatchNorm2d-183 [-1, 256, 14, 14] 512
ReLU-184 [-1, 256, 14, 14] 0
Conv2d-185 [-1, 256, 14, 14] 589,824
BatchNorm2d-186 [-1, 256, 14, 14] 512
ReLU-187 [-1, 256, 14, 14] 0
BasicBlock-188 [-1, 256, 14, 14] 0
Conv2d-189 [-1, 512, 7, 7] 1,179,648
BatchNorm2d-190 [-1, 512, 7, 7] 1,024
ReLU-191 [-1, 512, 7, 7] 0
Conv2d-192 [-1, 512, 7, 7] 2,359,296
BatchNorm2d-193 [-1, 512, 7, 7] 1,024
Conv2d-194 [-1, 512, 7, 7] 131,072
BatchNorm2d-195 [-1, 512, 7, 7] 1,024
ReLU-196 [-1, 512, 7, 7] 0
BasicBlock-197 [-1, 512, 7, 7] 0
Conv2d-198 [-1, 512, 7, 7] 2,359,296
BatchNorm2d-199 [-1, 512, 7, 7] 1,024
ReLU-200 [-1, 512, 7, 7] 0
Conv2d-201 [-1, 512, 7, 7] 2,359,296
BatchNorm2d-202 [-1, 512, 7, 7] 1,024
ReLU-203 [-1, 512, 7, 7] 0
BasicBlock-204 [-1, 512, 7, 7] 0
AdaptiveAvgPool2d-205 [-1, 512, 1, 1] 0
Identity-206 [-1, 512] 0
ResNet-207 [-1, 512] 0
Linear-208 [-1, 2] 3,074
================================================================
Total params: 33,532,610
Trainable params: 33,532,610
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 4096.00
Forward/backward pass size (MB): 188.38
Params size (MB): 127.92
Estimated Total Size (MB): 4412.30
----------------------------------------------------------------
