Transfer learning
实际工作中，只有很少的人从头开始训练 CNN，因为很难获得大量的样本。一般情况下，会通过调用预训练模型，例如
ConvNet在ImageNet（1.2 M图像1000个类别）,可以用ConvNet初始化，也可以作为特征提取器 用于感兴趣的任务或领域。
有两种主要的迁移学习：
- 微调
Convnet: 代替随机初始化，用预训练的网络模型进行初始化，就像在ImageNet1000的数据集上训练，剩下的训练就很常见了。 ConvNet作为固定的特征提取器: 冻结网络的权重，除了最后的全连接层，最后的全连接层被替换新的随机权重，然后只训练这一层。
加载数据
我们利用 torchvision 和 torch.utils.data 包加载数据。
今天的任务是训练 ants 和 bees 分类器模型，每个类别 120 张训练图，75 张图验证图，
数据集下载地址
Finetune
from __future__ import print_function, division
import torch
import torch.nn as nn
import torch.optim as optim
from torch.optim import lr_scheduler
import torch.backends.cudnn as cudnn
import numpy as np
import torchvision
from torchvision import datasets, models, transforms
import matplotlib.pyplot as plt
import time
import os
import copy
cudnn.benchmark = True
plt.ion() # interactive mode
# 对训练数据进行增强和归一化
# 对验证数据做归一化
data_transforms = {
'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),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]),
]),
}
# dataset and dataloader
data_dir = "../../../datasets/hymenoptera_data"
image_datasets = {
x: datasets.ImageFolder(os.path.join(data_dir, x), data_transforms[x])
for x in ['train', 'val']}
dataloaders = {
x: torch.utils.data.DataLoader(image_datasets[x], batch_size=4, shuffle=True,
num_workers=4)
for x in ['train', 'val']}
dataset_sizes = {x: len(image_datasets[x]) for x in ['train', 'val']}
class_names = image_datasets['train'].classes
device = torch.device(
"cuda") if torch.cuda.is_available() else torch.device("cpu")
# Visualize a few images
# Let’s visualize a few training images so as to understand the data augmentations.
#
def imshow(inp, title=None):
inp = inp.numpy().transpose((1, 2, 0))
mean = np.array([0.485, 0.456, 0.406])
std = np.array([0.229, 0.224, 0.225])
inp = std * inp + mean
inp = np.clip(inp, 0, 1)
plt.imshow(inp)
if title is not None:
plt.title(title)
plt.pause(0.001)
# inputs, classes = next(iter(dataloaders['train']))
# # make a grid from batch
# out = torchvision.utils.make_grid((inputs))
# imshow(out, title=[class_names[x] for x in classes])
# training
def train_model(model, criterion, optimizer, scheduler, num_epochs=25):
since = time.time()
best_model_wts = copy.deepcopy(model.state_dict())
best_acc = 0.0
for epoch in range(num_epochs):
print('Epoch {}/{}'.format(epoch, num_epochs-1))
print('-' * 10)
# Each epoch has a training and validation phase
for phase in ['train', 'val']:
if phase == 'train':
model.train() # Set model to training mode
else:
model.eval() # Set model to evaluate mode
running_loss = 0.0
running_corrects = 0
# Iterate over data.
for inputs, labels in dataloaders[phase]:
inputs = inputs.to(device)
labels = labels.to(device)
# zero the parameter gradients
optimizer.zero_grad()
# forward
# track history if only in train
with torch.set_grad_enabled(phase == 'train'):
outputs = model(inputs)
_, preds = torch.max(outputs, 1)
loss = criterion(outputs, labels)
# backward + optimize only if in training phase
if phase == 'train':
loss.backward()
optimizer.step()
# statisitc
running_loss += loss.item() * inputs.size(0)
running_corrects += torch.sum(preds == labels.data)
if phase == 'train':
scheduler.step()
epoch_loss = running_loss / dataset_sizes[phase]
epoch_acc = running_corrects.double() / dataset_sizes[phase]
print(f'{phase} Loss: {epoch_loss:.4f} Acc: {epoch_acc:.4f}')
# deep copy the model
if phase == 'val' and epoch_acc > best_acc:
best_acc = epoch_acc
best_model_wts = copy.deepcopy(model.state_dict())
print()
time_elapsed = time.time() - since
print(
f'Traing complete in {time_elapsed // 60:.0f}m {time_elapsed % 60:.0f} s')
print(f'Best val Acc: {best_acc:.4f}')
# load best model weights
model.load_state_dict(best_model_wts)
return model
# Generic function to display predictions for a few images
def visualize_model(model, num_images=6):
was_training = model.training
model.eval()
images_so_far = 0
fig = plt.figure()
with torch.no_grad():
for i, (inputs, labels) in enumerate(dataloaders['val']):
inputs = inputs.to(device)
labels = labels.to(device)
outputs = model(inputs)
_, preds = torch.max(outputs, 1)
for j in range(inputs.size()[0]):
images_so_far += 1
ax = plt.subplot(num_images // 2, 2, images_so_far)
ax.axis('off')
ax.set_title(f'predicted: {class_names[preds[j]]}')
imshow(inputs.cpu().data[j])
if images_so_far == num_images:
model.train(mode=was_training)
return
model.train(model=was_training)
# Load a pretrained model and reset final fully connected layer.
model_finetune = models.resnet18(pretrained=True)
num_ftrs = model_finetune.fc.in_features
# here the size of each output sample is set to 2
# it can be generalized to nn.Linear
model_finetune.fc = nn.Linear(num_ftrs, 2)
model_finetune = model_finetune.to(device)
criterion = nn.CrossEntropyLoss()
# Observe that all parameters are being optimized
optimizer_finetune = optim.SGD(
model_finetune.parameters(), lr=0.001, momentum=0.9)
# Decay LR by a factor of 0.1 every 7 epochs
exp_lr_scheduler = lr_scheduler.StepLR(
optimizer_finetune, step_size=7, gamma=0.1)
# train and evaluate
model_finetune = train_model(model=model_finetune,
criterion=criterion, optimizer=optimizer_finetune,
scheduler=exp_lr_scheduler, num_epochs=25)
# 可视化模型效果
visualize_model(model_finetune)
Output exceeds the size limit. Open the full output data in a text editor
train Loss: 0.6425 Acc: 0.6230
val Loss: 0.2763 Acc: 0.9150
Epoch 1/24
----------
train Loss: 0.4966 Acc: 0.8074
val Loss: 0.3128 Acc: 0.8693
Epoch 2/24
----------
train Loss: 0.4395 Acc: 0.8402
val Loss: 0.2923 Acc: 0.8562
Epoch 3/24
----------
train Loss: 0.3743 Acc: 0.8402
val Loss: 0.2049 Acc: 0.9281
Epoch 4/24
----------
train Loss: 0.2655 Acc: 0.8852
val Loss: 0.1841 Acc: 0.9346
Epoch 5/24
----------
...
val Loss: 0.1862 Acc: 0.9412
Traing complete in 1m 0 s
Best val Acc: 0.9477
固定为特征提取器
现在我们来冻结网络，除了最后一层。我们需要设置 requires_grad = False 来冻结参数，这样的话就不会计算梯度。查看更多
关键代码
# set requires_grad = False
for param in model_conv.parameters():
param.requires_grad = False
from __future__ import print_function, division
import torch
import torch.nn as nn
import torch.optim as optim
from torch.optim import lr_scheduler
import torch.backends.cudnn as cudnn
import numpy as np
import torchvision
from torchvision import datasets, models, transforms
import matplotlib.pyplot as plt
import time
import os
import copy
cudnn.benchmark = True
plt.ion() # interactive mode
# 对训练数据进行增强和归一化
# 对验证数据做归一化
data_transforms = {
'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),
transforms.ToTensor(),
transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]),
]),
}
# dataset and dataloader
data_dir = "../../../datasets/hymenoptera_data"
image_datasets = {
x: datasets.ImageFolder(os.path.join(data_dir, x), data_transforms[x])
for x in ['train', 'val']}
dataloaders = {
x: torch.utils.data.DataLoader(image_datasets[x], batch_size=4, shuffle=True,
num_workers=4)
for x in ['train', 'val']}
dataset_sizes = {x: len(image_datasets[x]) for x in ['train', 'val']}
class_names = image_datasets['train'].classes
device = torch.device(
"cuda") if torch.cuda.is_available() else torch.device("cpu")
# Visualize a few images
# Let’s visualize a few training images so as to understand the data augmentations.
#
def imshow(inp, title=None):
inp = inp.numpy().transpose((1, 2, 0))
mean = np.array([0.485, 0.456, 0.406])
std = np.array([0.229, 0.224, 0.225])
inp = std * inp + mean
inp = np.clip(inp, 0, 1)
plt.imshow(inp)
if title is not None:
plt.title(title)
plt.pause(0.001)
# inputs, classes = next(iter(dataloaders['train']))
# # make a grid from batch
# out = torchvision.utils.make_grid((inputs))
# imshow(out, title=[class_names[x] for x in classes])
# training
def train_model(model, criterion, optimizer, scheduler, num_epochs=25):
since = time.time()
best_model_wts = copy.deepcopy(model.state_dict())
best_acc = 0.0
for epoch in range(num_epochs):
print('Epoch {}/{}'.format(epoch, num_epochs-1))
print('-' * 10)
# Each epoch has a training and validation phase
for phase in ['train', 'val']:
if phase == 'train':
model.train() # Set model to training mode
else:
model.eval() # Set model to evaluate mode
running_loss = 0.0
running_corrects = 0
# Iterate over data.
for inputs, labels in dataloaders[phase]:
inputs = inputs.to(device)
labels = labels.to(device)
# zero the parameter gradients
optimizer.zero_grad()
# forward
# track history if only in train
with torch.set_grad_enabled(phase == 'train'):
outputs = model(inputs)
_, preds = torch.max(outputs, 1)
loss = criterion(outputs, labels)
# backward + optimize only if in training phase
if phase == 'train':
loss.backward()
optimizer.step()
# statisitc
running_loss += loss.item() * inputs.size(0)
running_corrects += torch.sum(preds == labels.data)
if phase == 'train':
scheduler.step()
epoch_loss = running_loss / dataset_sizes[phase]
epoch_acc = running_corrects.double() / dataset_sizes[phase]
print(f'{phase} Loss: {epoch_loss:.4f} Acc: {epoch_acc:.4f}')
# deep copy the model
if phase == 'val' and epoch_acc > best_acc:
best_acc = epoch_acc
best_model_wts = copy.deepcopy(model.state_dict())
print()
time_elapsed = time.time() - since
print(
f'Traing complete in {time_elapsed // 60:.0f}m {time_elapsed % 60:.0f} s')
print(f'Best val Acc: {best_acc:.4f}')
# load best model weights
model.load_state_dict(best_model_wts)
return model
# Generic function to display predictions for a few images
def visualize_model(model, num_images=6):
was_training = model.training
model.eval()
images_so_far = 0
fig = plt.figure()
with torch.no_grad():
for i, (inputs, labels) in enumerate(dataloaders['val']):
inputs = inputs.to(device)
labels = labels.to(device)
outputs = model(inputs)
_, preds = torch.max(outputs, 1)
for j in range(inputs.size()[0]):
images_so_far += 1
ax = plt.subplot(num_images // 2, 2, images_so_far)
ax.axis('off')
ax.set_title(f'predicted: {class_names[preds[j]]}')
imshow(inputs.cpu().data[j])
if images_so_far == num_images:
model.train(mode=was_training)
return
model.train(model=was_training)
### 不同的地方
# Load a pretrained model and reset final fully connected layer.
# model_finetune = models.resnet18(pretrained=True)
# num_ftrs = model_finetune.fc.in_features
model_conv = models.resnet18(pretrained=True)
# set requires_grad = False
for param in model_conv.parameters():
param.requires_grad = False
# 新的结构模块默认 requires_grad=True
num_ftrs = model_conv.fc.in_features
# here the size of each output sample is set to 2
# it can be generalized to nn.Linear
model_conv.fc = nn.Linear(num_ftrs, 2)
model_conv = model_conv.to(device)
criterion = nn.CrossEntropyLoss()
# Observe that all parameters are being optimized
optimizer_finetune = optim.SGD(
model_conv.parameters(), lr=0.001, momentum=0.9)
# Decay LR by a factor of 0.1 every 7 epochs
exp_lr_scheduler = lr_scheduler.StepLR(
optimizer_finetune, step_size=7, gamma=0.1)
# train and evaluate
model_conv = train_model(model=model_conv,
criterion=criterion, optimizer=optimizer_finetune,
scheduler=exp_lr_scheduler, num_epochs=25)
visualize_model(model_conv)
Output exceeds the size limit. Open the full output data in a text editor
Epoch 0/24
----------
train Loss: 0.6819 Acc: 0.6270
val Loss: 0.1885 Acc: 0.9477
Epoch 1/24
----------
train Loss: 0.4212 Acc: 0.7992
val Loss: 0.1760 Acc: 0.9477
Epoch 2/24
----------
train Loss: 0.5423 Acc: 0.7828
val Loss: 0.7761 Acc: 0.7059
Epoch 3/24
----------
train Loss: 0.6406 Acc: 0.7377
val Loss: 0.2656 Acc: 0.9216
Epoch 4/24
----------
train Loss: 0.5401 Acc: 0.7951
val Loss: 0.2485 Acc: 0.8954
...
val Loss: 0.1679 Acc: 0.9542
Traing complete in 0m 45 s
Best val Acc: 0.9542
【参考】
TRANSFER LEARNING FOR COMPUTER VISION TUTORIAL
