孪生网络pytoch实现，以resnet为特征提取网络
我的孪生网络代码来源于孪生网络博主，这里的源代码使用的vgg16作为特征提取网络，我的主要工作是将vgg16替换为resnet网络。
1.建立resnet网络
import torch.nn as nn
import torch
# from torchvision.models.utils import load_state_dict_from_url
from torchsummary import summary
from torch.hub import load_state_dict_from_url
class BasicBlock(nn.Module):#resnet18和resnet34的主干网络搭建
expansion = 1
def __init__(self, in_channel, out_channel, stride=1, downsample=None, **kwargs):
super(BasicBlock, self).__init__()
self.conv1 = nn.Conv2d(in_channels=in_channel, out_channels=out_channel,
kernel_size=3, stride=stride, padding=1, bias=False)
self.bn1 = nn.BatchNorm2d(out_channel)
self.relu = nn.ReLU()
self.conv2 = nn.Conv2d(in_channels=out_channel, out_channels=out_channel,
kernel_size=3, stride=1, padding=1, bias=False)
self.bn2 = nn.BatchNorm2d(out_channel)
self.downsample = downsample
def forward(self, x):
identity = x
if self.downsample is not None:
identity = self.downsample(x)
out = self.conv1(x)
out = self.bn1(out)
out = self.relu(out)
out = self.conv2(out)
out = self.bn2(out)
out += identity
out = self.relu(out)
return out
class Bottleneck(nn.Module):
"""
注意：原论文中，在虚线残差结构的主分支上，第一个1x1卷积层的步距是2，第二个3x3卷积层步距是1。
但在pytorch官方实现过程中是第一个1x1卷积层的步距是1，第二个3x3卷积层步距是2，
这么做的好处是能够在top1上提升大概0.5%的准确率。
可参考Resnet v1.5 https://ngc.nvidia.com/catalog/model-scripts/nvidia:resnet_50_v1_5_for_pytorch
"""
expansion = 4
def __init__(self, in_channel, out_channel, stride=1, downsample=None,
groups=1, width_per_group=64):
super(Bottleneck, self).__init__()
width = int(out_channel * (width_per_group / 64.)) * groups
self.conv1 = nn.Conv2d(in_channels=in_channel, out_channels=width,
kernel_size=1, stride=1, bias=False) # squeeze channels
self.bn1 = nn.BatchNorm2d(width)
# -----------------------------------------
self.conv2 = nn.Conv2d(in_channels=width, out_channels=width, groups=groups,
kernel_size=3, stride=stride, bias=False, padding=1)
self.bn2 = nn.BatchNorm2d(width)
# -----------------------------------------
self.conv3 = nn.Conv2d(in_channels=width, out_channels=out_channel*self.expansion,
kernel_size=1, stride=1, bias=False) # unsqueeze channels
self.bn3 = nn.BatchNorm2d(out_channel*self.expansion)
self.relu = nn.ReLU(inplace=True)
self.downsample = downsample
def forward(self, x):
identity = x
if self.downsample is not None:
identity = self.downsample(x)
out = self.conv1(x)
out = self.bn1(out)
out = self.relu(out)
out = self.conv2(out)
out = self.bn2(out)
out = self.relu(out)
out = self.conv3(out)
out = self.bn3(out)
out += identity
out = self.relu(out)
return out
class ResNet(nn.Module):
def __init__(self,block,blocks_num,num_classes=1000,include_top=True,groups=1,width_per_group=64):
super(ResNet, self).__init__()
self.include_top = include_top
self.in_channel = 64
self.groups = groups
self.width_per_group = width_per_group
self.conv1 = nn.Conv2d(3, self.in_channel, kernel_size=7, stride=2,padding=3, bias=False)
self.bn1 = nn.BatchNorm2d(self.in_channel)
self.relu = nn.ReLU(inplace=True)
self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
self.layer1 = self._make_layer(block, 64, blocks_num[0])
self.layer2 = self._make_layer(block, 128, blocks_num[1], stride=2)
self.layer3 = self._make_layer(block, 256, blocks_num[2], stride=2)
self.layer4 = self._make_layer(block, 512, blocks_num[3], stride=2)
if self.include_top:
self.avgpool = nn.AdaptiveAvgPool2d((1, 1)) # output size = (1, 1)
self.fc = nn.Linear(512 * block.expansion, num_classes)
for m in self.modules():
if isinstance(m, nn.Conv2d):
nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
def _make_layer(self, block, channel, block_num, stride=1):
downsample = None
if stride != 1 or self.in_channel != channel * block.expansion:
downsample = nn.Sequential(
nn.Conv2d(self.in_channel, channel * block.expansion, kernel_size=1, stride=stride, bias=False),
nn.BatchNorm2d(channel * block.expansion)
)
layers = []
layers.append(block(self.in_channel,channel,downsample=downsample,stride=stride,groups=self.groups,width_per_group=self.width_per_group))
self.in_channel = channel * block.expansion
for _ in range(1, block_num):
layers.append(block(self.in_channel,channel,groups=self.groups,width_per_group=self.width_per_group))
return nn.Sequential(*layers)
def forward(self, x):
x = self.conv1(x)
x = self.bn1(x)
x = self.relu(x)
x = self.maxpool(x)
x = self.layer1(x)
x = self.layer2(x)
x = self.layer3(x)
x = self.layer4(x)
if self.include_top:
x = self.avgpool(x)
x = torch.flatten(x, 1)
x = self.fc(x)
return x
def resnet34(pretrained,num_classes=1000, include_top=True):
# https://download.pytorch.org/models/resnet34-333f7ec4.pth
model = ResNet(BasicBlock, [3, 4, 6, 3], num_classes=num_classes, include_top=include_top)
if pretrained:
state_dict = load_state_dict_from_url("https://download.pytorch.org/models/resnet34-333f7ec4.pth",
model_dir="./model_data")
model.load_state_dict(state_dict)
return model
def resnet50(pretrained,num_classes=1000, include_top=True):
# https://download.pytorch.org/models/resnet50-19c8e357.pth
model=ResNet(Bottleneck, [3, 4, 6, 3], num_classes=num_classes, include_top=include_top)
if pretrained:
state_dict = load_state_dict_from_url("https://download.pytorch.org/models/resnet50-19c8e357.pth",
model_dir="./model_data")
model.load_state_dict(state_dict)
return model
def resnet101(pretrained,num_classes=1000, include_top=True):
# https://download.pytorch.org/models/resnet101-5d3b4d8f.pth
model=ResNet(Bottleneck, [3, 4, 23, 3], num_classes=num_classes, include_top=include_top)
if pretrained:
state_dict = load_state_dict_from_url("https://download.pytorch.org/models/resnet101-5d3b4d8f.pth",
model_dir="./model_data")
model.load_state_dict(state_dict)
return model
def resnext50_32x4d(num_classes=1000, include_top=True):
# https://download.pytorch.org/models/resnext50_32x4d-7cdf4587.pth
groups = 32
width_per_group = 4
return ResNet(Bottleneck, [3, 4, 6, 3],
num_classes=num_classes,
include_top=include_top,
groups=groups,
width_per_group=width_per_group)
def resnext101_32x8d(num_classes=1000, include_top=True):
# https://download.pytorch.org/models/resnext101_32x8d-8ba56ff5.pth
groups = 32
width_per_group = 8
return ResNet(Bottleneck, [3, 4, 23, 3],
num_classes=num_classes,
include_top=include_top,
groups=groups,
width_per_group=width_per_group)
if __name__=='__main__':
net=resnet101(False)
# del net.avgpool
summary(net,(3,105,105))
x=torch.rand(1,3,105,105)
out=net(x)
print(out.shape)2.修改调用网络的siamese.py文件
修改的主要思路就是按照原来调用vgg16的思路，将调用vgg16的地方全部修改为调用resnet网络。这里可以修改为调用resnet34，resnet50和resnet101.
原来的调用vgg16部分：
class Siamese(nn.Module):
def __init__(self, input_shape, pretrained=False):
super(Siamese, self).__init__()
self.vgg = VGG16(pretrained, input_shape[-1])
del self.vgg.avgpool
del self.vgg.classifier
flat_shape = 512 * get_img_output_length(input_shape[1], input_shape[0])
self.fully_connect1 = torch.nn.Linear(flat_shape, 512)
self.fully_connect2 = torch.nn.Linear(512, 1)
def forward(self, x):
x1, x2 = x
#------------------------------------------#
# 我们将两个输入传入到主干特征提取网络
#------------------------------------------#
x1 = self.vgg.features(x1)
x2 = self.vgg.features(x2)
#-------------------------#
# 相减取绝对值
#-------------------------#
x1 = torch.flatten(x1, 1)
x2 = torch.flatten(x2, 1)
x = torch.abs(x1 - x2)
#-------------------------#
# 进行两次全连接
#-------------------------#
x = self.fully_connect1(x)
x = self.fully_connect2(x)
return x修改后调用resnet网络的部分：这是调用resnet50的修改部分。
class Siamese(nn.Module):
def __init__(self, input_shape, pretrained=False):
super(Siamese, self).__init__()
self.resnet= resnet50(pretrained, include_top=True)
del self.resnet.avgpool
del self.resnet.fc
# flat_shape = 512 * get_img_output_length(input_shape[1], input_shape[0])
flat_shape = 2048 * 4*4
self.fully_connect1 = torch.nn.Linear(flat_shape, 512)
self.fully_connect2 = torch.nn.Linear(512, 1)
def forward(self, x):
x1, x2 = x
#------------------------------------------#
# 我们将两个输入传入到主干特征提取网络
#------------------------------------------#
x1 = self.resnet.conv1(x1)
x1 = self.resnet.bn1(x1)
x1 = self.resnet.relu(x1)
x1 = self.resnet.maxpool(x1)
x1 = self.resnet.layer1(x1)
x1 = self.resnet.layer2(x1)
x1 = self.resnet.layer3(x1)
x1 = self.resnet.layer4(x1)
x2 = self.resnet.conv1(x2)
x2 = self.resnet.bn1(x2)
x2 = self.resnet.relu(x2)
x2 = self.resnet.maxpool(x2)
x2 = self.resnet.layer1(x2)
x2 = self.resnet.layer2(x2)
x2 = self.resnet.layer3(x2)
x2 = self.resnet.layer4(x2)
#-------------------------#
# 相减取绝对值
#-------------------------#
x1 = torch.flatten(x1, 1)
x2 = torch.flatten(x2, 1)
x = torch.abs(x1 - x2)
#-------------------------#
# 进行两次全连接
#-------------------------#
x = self.fully_connect1(x)
x = self.fully_connect2(x)
return x对于原文的：
flat_shape = 512 * get_img_output_length(input_shape[1], input_shape[0])我不是很明白它的长宽是怎么计算的，我就根据resnet50的输出关系，将其直接手动修改为512*4*4，在resnet50的特征提取的最后一个池化层之后的输出特征维度就是512*4*4，当然这是对对应105*105长宽输入图像的时候，如果要自己修改一个图像输入尺寸，只需要看看resnet最后一个池化层的输出特征维度即可。如果是resnet101，则需要将这部分改为2048*4*4.
对于特征提取部分，源码写的是：
x1 = self.vgg.features(x1)
x2 = self.vgg.features(x2) 如果我按照源码这样去写是这样的：
x1 = self.resnet(x1)
x2 = self.resnet(x2) 但是这样总是会报错，因此我就将特征提取部分改为这样：
x1 = self.resnet.conv1(x1)
x1 = self.resnet.bn1(x1)
x1 = self.resnet.relu(x1)
x1 = self.resnet.maxpool(x1)
x1 = self.resnet.layer1(x1)
x1 = self.resnet.layer2(x1)
x1 = self.resnet.layer3(x1)
x1 = self.resnet.layer4(x1)
x2 = self.resnet.conv1(x2)
x2 = self.resnet.bn1(x2)
x2 = self.resnet.relu(x2)
x2 = self.resnet.maxpool(x2)
x2 = self.resnet.layer1(x2)
x2 = self.resnet.layer2(x2)
x2 = self.resnet.layer3(x2)
x2 = self.resnet.layer4(x2)对于最后的向量相似度计算：源码用的是相减的绝对值
x1 = torch.flatten(x1, 1)
x2 = torch.flatten(x2, 1)
x = torch.abs(x1 - x2)我试过：
x1 = torch.flatten(x1, 1)
x2 = torch.flatten(x2, 1)
x = torch.abs((x1 - x2)**2) x1 = torch.flatten(x1, 1)
x2 = torch.flatten(x2, 1)
x = torch.abs(x1*x1 - x2*x2) x1 = torch.flatten(x1, 1)
x2 = torch.flatten(x2, 1)
x = torch.abs((x1*x1 - x2*x2)**2)感觉结果都差不多。