1. GoogLeNet网络
1.1 整体网络结构
- 引入了Inception结构（融合不同尺度的特征信息）
- 使用1 x 1的卷积核进行降维以及映射处理
- 添加两个辅助分类器帮助训练
1.2 Inception网络结构
- 分支所得到的的特征矩阵高宽需要相同，才可以沿着深度方向拼接
- 1*1的卷积层可以起到降维的作用，可以减少计算量
3.辅助分类器
AlexNet和VGG只有一个输出，GooLeNet有三个输出。
第一层 AveragePool：平均池化下采样层，池化核大小为5×5，步距为3
从Inception(4a)中输入到该层时，输入特征矩阵大小为14×14×512，输出为4×4×512；
从Inception(4b)中输入到该层时，输入特征矩阵大小为14×14×528 ，输出为4×4×528；
第二层 Conv：128个卷积核大小为1×1的卷积层（使用了ReLU激活函数），降维
第三层 FC：节点个数为1024的全连接层（使用了ReLU激活函数）
两个全连接层之间使用了dropout函数（70%的比例随机失活神经元）
第四层 FC：全连接层，对于ImageNet数据集有1000个类别，节点个数就是1000
Softmax激活函数
2. 网络构建
2.1 BasicConv2d构建
因为Conv2d函数通常与一个ReLu函数一起用，故构建一个BasicConv2d函数用于卷积层。
class BasicConv2d(nn.Module):
def __init__(self,in_channels,out_channels,kernel_size):
super(BasicConv2d, self).__init__()
self.conv = nn.Conv2d(in_channels,out_channels,kernel_size)
self.relu = nn.ReLU(inplace=True)
def forward(self,x):
x = self.conv(x)
x = self.relu(x)
return x2.2 Inception构建
class Inception(nn.Module):
def __init__(self,in_channels,conv1x1,conv3x3reduce,conv3x3,conv5x5reduce,conv5x5,pool_project):
super(Inception, self).__init__()
self.branch1 = BasicConv2d(in_channels,conv1x1,kernel_size=1)
self.branch2 = nn.Sequential(
BasicConv2d(in_channels,conv3x3reduce,kernel_size=3),
BasicConv2d(conv3x3reduce,conv3x3,kernel_size=1)
)
self.branch3 = nn.Sequential(
BasicConv2d(in_channels, conv5x5reduce, kernel_size=3),
BasicConv2d(conv3x3reduce, conv5x5, kernel_size=1)
)
self.branch4 = nn.Sequential(
#该层池化后输入大小等于输出大小
nn.MaxPool2d(kernel_size=3,stride=1,padding=1),
nn.Conv2d(in_channels,pool_project,kernel_size=1)
)
def forward(self,x):
branch1 = self.branch1(x)
branch2 = self.branch2(x)
branch3 = self.branch3(x)
branch4 = self.branch4(x)
output = [branch1,branch2,branch3,branch4]
#在dim=1维度上进行合并，因为第0维度是batch
return torch.cat(output,dim=1)
2.3 辅助分类器的构建
#两个辅助分类器参数一样
import torch
from torch import nn
import torch.nn.functional as F
class Aux(nn.Module):
def __init__(self, in_channels, num_classes):
super(Aux, self).__init__()
self.averagePool = nn.AvgPool2d(kernel_size=5, stride=3) # output[batch, 512, 4, 4]/output[batch, 528, 4, 4]/
self.conv = BasicConv2d(in_channels, 128, kernel_size=1) # output[batch, 128, 4, 4]
#定义第二个卷积层BasicConv2d 输入特征矩阵深度就是我们输入的in_channels，卷积核个数是128，卷积核大小为1x1
# output batch不会改变 128是卷积核个数也是输出矩阵的深度128 4x4是输出的大小 是不会改变的 因为使用的卷积核大小是1x1的
self.fc1 = nn.Linear(2048, 1024)
self.fc2 = nn.Linear(1024, num_classes)
def forward(self, x):
x = self.averagePool(x)
x = self.conv(x)
x = torch.flatten(x, 1)
x = F.dropout(x, 0.5, training=self.training)
# N x 2048
x = F.relu(self.fc1(x), inplace=True)
x = F.dropout(x, 0.5, training=self.training)
x = self.fc2(x)
return x
此处注意dropuout的用法，nn.dropout不能输入参数，需要定义在函数里面
2.4 构建model
class GoogLeNet(nn.Module):
def __init__(self,num_class = 1000,aux_logist = True,init_weight = False):
super(GoogLeNet, self).__init__()
self.aux_logits = aux_logist
# (224-7+2*3)/2+1=112.5,向下取整,输出112
self.conv1 = BasicConv2d(3, 64, kernel_size=7, stride=2, padding=3)
self.maxpool1 = nn.MaxPool2d(3, stride=2, ceil_mode=True) # ceil_mode=True 向上取整 False 向下取整
self.conv2 = BasicConv2d(64, 64, kernel_size=1)
self.conv3 = BasicConv2d(64, 192, kernel_size=3, padding=1)
self.maxpool2 = nn.MaxPool2d(3, stride=2, ceil_mode=True)
self.inception3a = Inception(192, 64, 96, 128, 16, 32, 32)
self.inception3b = Inception(256, 128, 128, 192, 32, 96, 64)
self.maxpool3 = nn.MaxPool2d(3, stride=2, ceil_mode=True)
self.inception4a = Inception(480, 192, 96, 208, 16, 48, 64)
self.inception4b = Inception(512, 160, 112, 224, 24, 64, 64)
self.inception4c = Inception(512, 128, 128, 256, 24, 64, 64)
self.inception4d = Inception(512, 112, 144, 288, 32, 64, 64)
self.inception4e = Inception(528, 256, 160, 320, 32, 128, 128)
self.maxpool4 = nn.MaxPool2d(3, stride=2, ceil_mode=True)
self.inception5a = Inception(832, 256, 160, 320, 32, 128, 128)
self.inception5b = Inception(832, 384, 192, 384, 48, 128, 128)
if self.aux_logits: # 如果使用辅助分类器的话，这个是我们刚刚传入的布尔参数
aux = AUX()
self.aux1 = aux(512, num_class) # 输入特侦矩阵的深度（inception4a的输出）
self.aux2 = aux(528, num_class) # 输入特侦矩阵的深度（inception4d的输出）
self.avgpool = nn.AdaptiveAvgPool2d((1, 1)) # 自适应的平均池化下采样操作 参数所需要输出特征矩阵的高和宽
# 好处：无论输入的特征矩阵的高和宽是多少 最后的输出都可以固定为1x1
self.dropout = nn.Dropout(0.4)
self.fc = nn.Linear(1024, num_class)
# 输入的是展平后的向量节点个数是1024，输出节点个数是num_classes
if init_weight: # 如果有初始化权重参数就是用 这里的详解在AlexNet模型笔记中
self._initialize_weights()
def forward(self, x): # 定义下采样层
# N x 3 x 224 x 224
x = self.conv1(x)
# N x 64 x 112 x 112
x = self.maxpool1(x)
# N x 64 x 56 x 56
x = self.conv2(x)
# N x 64 x 56 x 56
x = self.conv3(x)
# N x 192 x 56 x 56
x = self.maxpool2(x)
# N x 192 x 28 x 28
x = self.inception3a(x)
# N x 256 x 28 x 28
x = self.inception3b(x)
# N x 480 x 28 x 28
x = self.maxpool3(x)
# N x 480 x 14 x 14
x = self.inception4a(x)
# N x 512 x 14 x 14
if self.training and self.aux_logits: # eval model lose this layer
# 第一个参数是判断我们的当前处于的训练状态是否在训练状态 然后是是否使用辅助分类器
aux1 = self.aux1(x)
x = self.inception4b(x)
# N x 512 x 14 x 14
x = self.inception4c(x)
# N x 512 x 14 x 14
x = self.inception4d(x)
# N x 528 x 14 x 14
if self.training and self.aux_logits: # eval model lose this layer
# 第一个参数是判断我们的当前处于的训练状态是否在训练状态 然后是是否使用辅助分类器
aux2 = self.aux2(x)
x = self.inception4e(x)
# N x 832 x 14 x 14
x = self.maxpool4(x)
# N x 832 x 7 x 7
x = self.inception5a(x)
# N x 832 x 7 x 7
x = self.inception5b(x)
# N x 1024 x 7 x 7
x = self.avgpool(x)
# N x 1024 x 1 x 1
x = torch.flatten(x, 1)
# N x 1024
x = self.dropout(x)
x = self.fc(x)
# N x 1000 (num_classes)
if self.training and self.aux_logits: # eval model lose this layer
return x, aux2, aux1
# 返回三个值 一个是我们的主输出值 辅助分类器2的值 辅助分类器1的值
return x
def _initialize_weights(self):
for m in self.modules():
if isinstance(m, nn.Conv2d):
nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
if m.bias is not None:
nn.init.constant_(m.bias, 0)
elif isinstance(m, nn.Linear):
nn.init.normal_(m.weight, 0, 0.01)
nn.init.constant_(m.bias, 0)
3.训练
与AlexNet，VGG相似，先加载数据
import json
import torch
from torch import optim
from torch.utils.data import DataLoader
from torchvision import datasets,transforms
from model import vgg
def main():
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(device)
data_transform = {
"train": transforms.Compose([transforms.RandomResizedCrop(224),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))]),
"val": transforms.Compose([transforms.Resize((224, 224)),
transforms.ToTensor(),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])}
train_path = "zhang/train"
val_path = "zhang/val"
train_data = datasets.ImageFolder(root=train_path,transform=data_transform["train"])
val_data = datasets.ImageFolder(root=val_path,transform = data_transform["val"])
train_dataset = DataLoader(train_data,batch_size=32,shuffle=True)
val_dataset = DataLoader(val_data, batch_size=32, shuffle=True)
val_num = len(val_data)
flower_list = train_data.class_to_idx
class_dict = dict((val, key) for key, val in flower_list.items()) # 将 key 和val 反过来
json_str = json.dumps(class_dict, indent=4) # indent=4换行，更加美观
with open("class_indices.json", "w", encoding="utf-8") as f:
f.write(json_str) net = GoogLeNet(num_class=5,aux_logist=True,init_weight=True)
net.to(device)
loss_function = torch.nn.CrossEntropyLoss()
optimizer = optim.Adam(net.parameters(), lr=0.0003)
best_acc = 0.0 # 用于后面设置准确率
for epoch in range(30):
print(f"----------第{epoch + 1}轮训练开始--------")
# train dropout使用后需要.train .eval
net.train()
for i, (images, labels) in enumerate(train_dataset):
optimizer.zero_grad()
output1, aux_output2, aux_output1 = net(images.to(device))
loss0 = loss_function(output1, labels.to(device))
loss1 = loss_function(aux_output1, labels.to(device))
loss2 = loss_function(aux_output2, labels.to(device))
loss = loss0 + loss1 * 0.3 + loss2 * 0.3
# 优化模型
loss.backward()
optimizer.step()
# val
net.eval()
accuracy = 0.0
with torch.no_grad():
for data in val_dataset:
test_Img, target = data
output = net(test_Img.to(device))
# dim=1表示的是行
predict_y = torch.argmax(output, dim=1)
accuracy += (predict_y == target.to(device)).sum().item()
acc_rate = accuracy / val_num
# 选取训练效果最好的模型保存
if acc_rate > best_acc:
best_acc = acc_rate
torch.save(net.state_dict(), './GoogLeNet.pth')
print(acc_rate)
if __name__ == '__main__':
main()与AlexNet的不同点
#将我们的一批数据 训练图像传入到我们的网络当中，会得到三个输出
#一个主输出，两个辅助分类器的输出
loss = loss0 + loss1 * 0.3 + loss2 * 0.3
#原论文中就是以0.3的权重配置到损失当中
#之前网络当中只有一个输出 现在采用了两个辅助分类器 ，所以一共有三个输出
4.predict
需要注意构建模型的时候，载入模型的部分不需要构建辅助分类器aux_logits=False
但是我们保存模型的时候，已经将我们的参数保存在我们的模型当中，所以我们要将我们的strict参数设为False
import json
import torch
from PIL import Image
from torchvision import transforms
from model import GoogLeNet
data_transform = transforms.Compose([
transforms.Resize((224,244)),
transforms.ToTensor(),
transforms.Normalize((0.5,0.5,0.5),(0.5,0.5,0.5))
])
img_path = "./th.jpg"
img = Image.open(img_path)
img = data_transform(img)
#扩充维度，否则图片是三维的
img = torch.unsqueeze(img,dim = 0)
try:
json_file = open("class_indices.json","r")
class_indict = json.load(json_file)
except Exception as e:
print(e)
exit(-1)
#初始化网络
model = GoogLeNet(num_class=5, aux_logist=False)
model_path = "GoogLeNet.pth"
missing_keys, unexpected_keys = model.load_state_dict(torch.load(model_path), strict=False)
#在unexpe_keys会有一系列层 这些层都是属于那些辅助分类器的
model.eval()
with torch.no_grad():
output = torch.squeeze(model(img)) #压缩掉第一个维度
predict = torch.softmax(output,dim=0) #得到概率
predict_cla = torch.argmax(predict).numpy()
print(class_indict[str(predict_cla)],predict[predict_cla].item())
扫一扫
746
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
