1. ResNet网络
1.1 整体网络结构
论文:《Deep Residual Learning for Image Recognition》
网络中的亮点:
- 超深的网络结构。(突破1000层)
- 提出 Residual 模块。
- 使用 Batch Normalization 加速训练。(丢弃dropout)
对于网络模型不断加深,会出现的问题:①梯度消失和梯度爆炸现象②退化现象
解决方法 :①对数据进行标准化处理、还有权重初始化,或者是BN初始化处理,可以解决梯度消失和梯度爆炸。
②Residual 模块解决退化现象。
1.2 Batch Normalization
BN:Batch Normalization
- 计算一个Batch数据的feature map然后在进行标准化(batch越大越接近整个数据集的分布,效果越好)
- BN层放在卷积层和激活层之间,且无需设置bias
- 训练时需要设置True,测试时设置False
1.3 residual 结构
residual 结构
主分支和短路连接(Shortcut)结果相加
残差模块:一条路不变(恒等映射);另一条路负责拟合相对于原始网络的残差,去纠正原始网络的偏差,而不是让整体网络去拟合全部的底层映射,这样网络只需要纠正偏差。
虚线表示的是用1x1的卷积核进行了维度处理,用于深层的神经网络。
1.4 分组卷积
将一个图像的channel拆分成不同的group卷积,这样所需要的参数更少(3通道以上使用)
2. 网络构建
2.1 构建残差网络
# 18或34层的残差结构。
class BasicBlock(nn.Module):
expansion = 1 # 残差结构所使用卷积核个数的一个变化,1表示是之前的一倍,没有变化。
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) # BatchNorm2d不需要偏置。
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 # downsample为是虚线还是实线的残差结构。
def forward(self, x):
identity = x # 残差结构的分支
if self.downsample is not None: # 是否是虚线的残差结构
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
# 50、101、152层的残差结构。
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 # 卷积核个数变为原来的4倍
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, # 这里卷积核个数变为原来的4倍。
kernel_size=1, stride=1, bias=False) # unsqueeze channels
self.bn3 = nn.BatchNorm2d(out_channel*self.expansion)
self.relu = nn.ReLU(inplace=True) # inplace为True不影响结果,可以节省内存。
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
2.2 创造各个卷积层
# 创建每种卷积层。
# channel为主分支卷积核的个数,block_num为卷积层的层数。
def _make_layer(self, block, channel, block_num, stride=1):
downsample = None
# 分支是否是虚线。
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 = [] # 新建一个列表,存放每一个卷积层。
# 第一层的分支为虚线,先添加进去。
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 # 更新输入通道数
# 遍历剩下的层,逐个添加到列表中。
for _ in range(1, block_num): # 从1开始,因为第一层已经添加了,0表示第一层。
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):
# conv1
x = self.conv1(x)
x = self.bn1(x)
x = self.relu(x)
x = self.maxpool(x)
x = self.layer1(x) # conv2_x
x = self.layer2(x) # conv3_x
x = self.layer3(x) # conv4_x
x = self.layer4(x) # conv5_x
if self.include_top:
x = self.avgpool(x) #池化
x = torch.flatten(x, 1) # 展平处理
x = self.fc(x) #全联接层
return x
def forward(self, x):
# conv1
x = self.conv1(x)
x = self.bn1(x)
x = self.relu(x)
x = self.maxpool(x)
x = self.layer1(x) # conv2_x
x = self.layer2(x) # conv3_x
x = self.layer3(x) # conv4_x
x = self.layer4(x) # conv5_x
if self.include_top:
x = self.avgpool(x) #池化
x = torch.flatten(x, 1) # 展平处理
x = self.fc(x) #全联接层
return x
2.3 定义网络模型
# 定义ResNet模型。
class ResNet(nn.Module):
def __init__(self,
block, # 用的那种残差结构
blocks_num, # 每层的残差结构的个数
num_classes=1000, # 类别数目
include_top=True, # 表示在ResNet的基础上构建更复杂的模型,默认为true
groups=1,
width_per_group=64):
super(ResNet, self).__init__()
self.include_top = include_top
self.in_channel = 64 # 输入的通道数为64
self.groups = groups
self.width_per_group = width_per_group
# 第一种卷积层 conv1。
# RGD图片输入的通道数为3,输出为64(卷积核的组数)。
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]) # 对应conv2_x
self.layer2 = self._make_layer(block, 128, blocks_num[1], stride=2) # 对应conv3_x
self.layer3 = self._make_layer(block, 256, blocks_num[2], stride=2) # 对应conv4_x
self.layer4 = self._make_layer(block, 512, blocks_num[3], stride=2) # 对应conv5_x
if self.include_top:
# 自适应采样,不管输入尺寸是多少,输出为(1, 1)。
self.avgpool = nn.AdaptiveAvgPool2d((1, 1)) # output size = (1, 1)
self.fc = nn.Linear(512 * block.expansion, num_classes) # 全连接层
# 初始化权重。
for m in self.modules():
if isinstance(m, nn.Conv2d):
nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
3. Train
- 加载预训练模型的方法
model_weight_path = "resnet34.pth" # 预训练模型的权重
net.load_state_dict(torch.load(model_weight_path, map_location='cpu'))- 对batch进行归一化
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])])- 冻结网络,只训练最后一层Linear
for param in net.parameters():
param.requires_grad = False
in_channel = net.fc.in_features # 全连接层输入的深度。
net.fc = nn.Linear(in_channel, 5)
import json
import torch
import torch.nn as nn
import torch.optim as optim
from torch.utils.data import DataLoader
from torchvision import transforms, datasets
from model import resnet34
def main():
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
print(device)
data_transform = {
"train": transforms.Compose([transforms.RandomResizedCrop(224),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])]),
"val": transforms.Compose([transforms.Resize(256),
transforms.CenterCrop(224), # 中心裁剪
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])])}
train_path = "F:/HUWEI/pytorch_learning/GoogLeNet/zhang/train"
val_path = "F:/HUWEI/pytorch_learning/GoogLeNet/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)
# {'daisy':0, 'dandelion':1, 'roses':2, 'sunflower':3, 'tulips':4}
flower_list = train_data.class_to_idx
cla_dict = dict((val, key) for key, val in flower_list.items())
# write dict into json file
json_str = json.dumps(cla_dict, indent=4)
with open('class_indices.json', 'w') as json_file:
json_file.write(json_str)
# 初始化模型。
net = resnet34()
# 使用迁移学习的方式进行训练。
# load pretrain weights
# download url: https://download.pytorch.org/models/resnet34-333f7ec4.pth
model_weight_path = "resnet34-333f7ec4.pth" # 预训练模型的权重
net.load_state_dict(torch.load(model_weight_path, map_location='cpu')) # 加载预训练模型
for param in net.parameters():
param.requires_grad = False
# change fc layer structure
in_channel = net.fc.in_features # 全连接层输入的深度。
net.fc = nn.Linear(in_channel, 5)
net.to(device)
# define loss function
loss_function = nn.CrossEntropyLoss()
# construct an optimizer
params = [p for p in net.parameters() if p.requires_grad]
optimizer = optim.Adam(params, lr=0.0001)
epochs = 3
best_acc = 0.0
save_path = './resNet34.pth'
train_steps = len(train_dataset)
for epoch in range(epochs):
# train
net.train()# 控制网络中BatchNorm2d的状态。
running_loss = 0.0
for step, data in enumerate(train_dataset):
images, labels = data
optimizer.zero_grad()
logits = net(images.to(device))
loss = loss_function(logits, labels.to(device))
loss.backward()
optimizer.step()
# print statistics
running_loss += loss.item()
# validate
net.eval() # 控制网络中BatchNorm2d的状态。
acc = 0.0 # accumulate accurate number / epoch
with torch.no_grad():
for val_data in val_dataset:
val_images, val_labels = val_data
outputs = net(val_images.to(device))
# loss = loss_function(outputs, test_labels)
predict_y = torch.max(outputs, dim=1)[1]
acc += torch.eq(predict_y, val_labels.to(device)).sum().item()
val_accurate = acc / val_num
print('[epoch %d] train_loss: %.3f val_accuracy: %.3f' %
(epoch + 1, running_loss / train_steps, val_accurate))
if val_accurate > best_acc:
best_acc = val_accurate
torch.save(net.state_dict(), save_path)
print('Finished Training')
if __name__ == '__main__':
main()
4. predict
import json
import torch
from PIL import Image
from torchvision import transforms
from model import resnet34
data_transform = transforms.Compose([
transforms.Resize((224,244)),
transforms.ToTensor(),
transforms.Normalize((0.485, 0.456, 0.406), (0.229, 0.224, 0.225))
])
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 = resnet34(num_classes=5)
model_path = "resNet34.pth"
model.load_state_dict(torch.load(model_path))
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())
需对batch归一化