pytorch之迁移学习
实际中，基本没有人会从零开始（随机初始化）训练一个完整的卷积网络，因为相对于网络，很难得到一个足够大的数据集[网络很深, 需要足够大数据集]。通常的做法是在一个很大的数据集上进行预训练得到卷积网络ConvNet, 然后将这个ConvNet的参数作为目标任务的初始化参数或者固定这些参数。
转移学习的两个主要场景：
- 微调Convnet：使用预训练的网络(如在imagenet 1000上训练而来的网络)来初始化自己的网络，而不是随机初始化。其他的训练步骤不变。
- 将Convnet看成固定的特征提取器:首先固定ConvNet除了最后的全连接层外的其他所有层。最后的全连接层被替换成一个新的随机 初始化的层，只有这个新的层会被训练[只有这层参数会在反向传播时更新]
下面是利用PyTorch进行迁移学习步骤，要解决的问题是训练一个模型来对蚂蚁和蜜蜂进行分类。
1.导入相关的包
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 numpy as np
import torchvision
from torchvision import datasets, models, transforms
import matplotlib.pyplot as plt
import time
import os
import copy
plt.ion() # 交互模式
注释1：交互模式详情
2.加载数据
今天要解决的问题是训练一个模型来分类蚂蚁ants和蜜蜂bees。ants和bees各有约120张训练图片。每个类有75张验证图片。从零开始在 如此小的数据集上进行训练通常是很难泛化的。由于我们使用迁移学习，模型的泛化能力会相当好。 该数据集是imagenet的一个非常小的子集。从此处下载数据，并将其解压缩到当前目录。
# 训练集数据扩充和归一化
# 在验证集上仅需要归一化
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])
]),
}
data_dir = 'data/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:0" if torch.cuda.is_available() else "cpu")
注释2：ImageFolder
torchvision中有一个更常用的数据集类ImageFolder。 它假定了数据集是以如下方式构造的:
root/ants/xxx.png
root/ants/xxy.jpeg
root/ants/xxz.png
.
.
.
root/bees/123.jpg
root/bees/nsdf3.png
root/bees/asd932_.png
也就是"根目录/类别名称/该类别对应的图片"
3.可视化部分图像数据
可视化部分训练图像，以便了解数据扩充。
def imshow(inp, title=None):
"""Imshow for Tensor."""
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) # pause a bit so that plots are updated
# 获得一批训练数据
inputs, classes = next(iter(dataloaders['train']))
# 批量制作网格
out = torchvision.utils.make_grid(inputs)
imshow(out, title=[class_names[x] for x in classes])
4.训练模型
编写一个通用函数来训练模型。下面将说明：
- 调整学习速率
- 保存最好的模型
下面的参数scheduler是一个来自 torch.optim.lr_scheduler的学习速率调整类的对象(LRscheduler object)。
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)
# 将参数梯度归零
optimizer.zero_grad()
# forward
# 只在训练上追踪历史
with torch.set_grad_enabled(phase == 'train'):
outputs = model(inputs)
_, preds = torch.max(outputs, 1)
loss = criterion(outputs, labels)
# backward + optimize 只在训练进行
if phase == 'train':
loss.backward()
optimizer.step()
# statistics
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('{} Loss: {:.4f} Acc: {:.4f}'.format(
phase, epoch_loss, epoch_acc))
# 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('Training complete in {:.0f}m {:.0f}s'.format(
time_elapsed // 60, time_elapsed % 60))
print('Best val Acc: {:4f}'.format(best_acc))
# load best model weights
model.load_state_dict(best_model_wts)
return model
5.可视化模型的预测结果
一个通用的展示少量预测图片的函数
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('predicted: {}'.format(class_names[preds[j]]))
imshow(inputs.cpu().data[j])
if images_so_far == num_images:
model.train(mode=was_training)
return
model.train(mode=was_training)
6.场景1：微调ConvNet
加载预训练模型并重置最终完全连接的图层。
model_ft = models.resnet18(pretrained=True)
# in_features 是fc线性层的输入数量
num_ftrs = model_ft.fc.in_features
# 这里，每个输出样本的大小设置为2。
# 或者，它可以推广到nn.Linear（num_ftrs，len（类名称））。
model_ft.fc = nn.Linear(num_ftrs, 2)
model_ft = model_ft.to(device)
criterion = nn.CrossEntropyLoss()
# 观察所有参数都正在优化
optimizer_ft = optim.SGD(model_ft.parameters(), lr=0.001, momentum=0.9)
# 每7个epochs衰减LR通过设置gamma=0.1
exp_lr_scheduler = lr_scheduler.StepLR(optimizer_ft, step_size=7, gamma=0.1)
训练和评估模型
（1）训练模型 该过程在CPU上需要大约15-25分钟，但是在GPU上，它只需不到一分钟。（我在CPU上跑的）
model_ft = train_model(model_ft, criterion, optimizer_ft, exp_lr_scheduler,
num_epochs=25)
- 输出
Epoch 0/24
----------
train Loss: 0.6633 Acc: 0.6721
val Loss: 0.2236 Acc: 0.9216
Epoch 1/24
----------
train Loss: 0.4831 Acc: 0.7705
val Loss: 0.2813 Acc: 0.9020
Epoch 2/24
----------
train Loss: 0.4444 Acc: 0.7828
val Loss: 0.1721 Acc: 0.9477
Epoch 3/24
----------
train Loss: 0.4857 Acc: 0.7828
val Loss: 0.1651 Acc: 0.9477
Epoch 4/24
----------
train Loss: 0.5882 Acc: 0.7705
val Loss: 0.4952 Acc: 0.8301
Epoch 5/24
----------
train Loss: 0.5504 Acc: 0.7869
val Loss: 0.1896 Acc: 0.9412
Epoch 6/24
----------
train Loss: 0.5927 Acc: 0.7992
val Loss: 0.3142 Acc: 0.8954
Epoch 7/24
----------
train Loss: 0.4391 Acc: 0.8361
val Loss: 0.1705 Acc: 0.9477
Epoch 8/24
----------
train Loss: 0.3733 Acc: 0.8566
val Loss: 0.1884 Acc: 0.9346
Epoch 9/24
----------
train Loss: 0.3567 Acc: 0.8484
val Loss: 0.2050 Acc: 0.9281
Epoch 10/24
----------
train Loss: 0.3769 Acc: 0.8279
val Loss: 0.2070 Acc: 0.9346
Epoch 11/24
----------
train Loss: 0.3473 Acc: 0.8648
val Loss: 0.2191 Acc: 0.9281
Epoch 12/24
----------
train Loss: 0.3654 Acc: 0.8566
val Loss: 0.1732 Acc: 0.9412
Epoch 13/24
----------
train Loss: 0.2885 Acc: 0.8689
val Loss: 0.1959 Acc: 0.9346
Epoch 14/24
----------
train Loss: 0.3242 Acc: 0.8525
val Loss: 0.2066 Acc: 0.9281
Epoch 15/24
----------
train Loss: 0.3471 Acc: 0.8279
val Loss: 0.1821 Acc: 0.9477
Epoch 16/24
----------
train Loss: 0.4058 Acc: 0.8443
val Loss: 0.1773 Acc: 0.9346
Epoch 17/24
----------
train Loss: 0.4398 Acc: 0.8279
val Loss: 0.1726 Acc: 0.9477
Epoch 18/24
----------
train Loss: 0.3293 Acc: 0.8689
val Loss: 0.1841 Acc: 0.9346
Epoch 19/24
----------
train Loss: 0.3484 Acc: 0.8361
val Loss: 0.1846 Acc: 0.9412
Epoch 20/24
----------
train Loss: 0.3164 Acc: 0.8402
val Loss: 0.1702 Acc: 0.9542
Epoch 21/24
----------
train Loss: 0.3769 Acc: 0.8197
val Loss: 0.1828 Acc: 0.9346
Epoch 22/24
----------
train Loss: 0.3204 Acc: 0.8852
val Loss: 0.2065 Acc: 0.9412
Epoch 23/24
----------
train Loss: 0.3201 Acc: 0.8852
val Loss: 0.1970 Acc: 0.9346
Epoch 24/24
----------
train Loss: 0.2603 Acc: 0.8730
val Loss: 0.2063 Acc: 0.9412
Training complete in 33m 18s
Best val Acc: 0.954248
Process finished with exit code 0
（2）模型评估结果可视化
visualize_model(model_ft)
- 输出
7.场景2：ConvNet作为固定特征提取器
在这里需要冻结除最后一层之外的所有网络。通过设置requires_grad == Falsebackward()来冻结参数，这样在反向传播backward()的时候他们的梯度就不会被计算。
model_conv = torchvision.models.resnet18(pretrained=True)
for param in model_conv.parameters():
param.requires_grad = False
# 默认情况下，新建模块的参数需要requires_grad=True
num_ftrs = model_conv.fc.in_features
model_conv.fc = nn.Linear(num_ftrs, 2)
model_conv = model_conv.to(device)
criterion = nn.CrossEntropyLoss()
# 请注意，只有最后一层的参数被优化为
# 与以前相反。
optimizer_conv = optim.SGD(model_conv.fc.parameters(), lr=0.001, momentum=0.9)
# 每7个epochs将LR衰减0.1倍
exp_lr_scheduler = lr_scheduler.StepLR(optimizer_conv, step_size=7, gamma=0.1)
训练和评估
（1）训练模型 在CPU上，与前一个场景相比，这将花费大约一半的时间，因为不需要为大多数网络计算梯度。但需要计算转发。
model_conv = train_model(model_conv, criterion, optimizer_conv,
exp_lr_scheduler, num_epochs=25)
- 输出
Epoch 0/24
----------
train Loss: 0.5987 Acc: 0.6393
val Loss: 0.3066 Acc: 0.8693
Epoch 1/24
----------
train Loss: 0.5367 Acc: 0.7172
val Loss: 0.1962 Acc: 0.9477
Epoch 2/24
----------
train Loss: 0.4591 Acc: 0.7992
val Loss: 0.1974 Acc: 0.9477
Epoch 3/24
----------
train Loss: 0.5040 Acc: 0.7787
val Loss: 0.1922 Acc: 0.9346
Epoch 4/24
----------
train Loss: 0.5343 Acc: 0.7705
val Loss: 0.2062 Acc: 0.9542
Epoch 5/24
----------
train Loss: 0.5048 Acc: 0.7828
val Loss: 0.2255 Acc: 0.9281
Epoch 6/24
----------
train Loss: 0.5591 Acc: 0.7951
val Loss: 0.2150 Acc: 0.9346
Epoch 7/24
----------
train Loss: 0.3710 Acc: 0.8402
val Loss: 0.2361 Acc: 0.9346
Epoch 8/24
----------
train Loss: 0.2645 Acc: 0.8934
val Loss: 0.2024 Acc: 0.9346
Epoch 9/24
----------
train Loss: 0.3999 Acc: 0.8402
val Loss: 0.1959 Acc: 0.9477
Epoch 10/24
----------
train Loss: 0.3806 Acc: 0.8238
val Loss: 0.2191 Acc: 0.9346
Epoch 11/24
----------
train Loss: 0.4044 Acc: 0.8402
val Loss: 0.1941 Acc: 0.9542
Epoch 12/24
----------
train Loss: 0.3234 Acc: 0.8648
val Loss: 0.1977 Acc: 0.9477
Epoch 13/24
----------
train Loss: 0.3640 Acc: 0.8361
val Loss: 0.2026 Acc: 0.9346
Epoch 14/24
----------
train Loss: 0.4070 Acc: 0.8115
val Loss: 0.1912 Acc: 0.9542
Epoch 15/24
----------
train Loss: 0.3331 Acc: 0.8484
val Loss: 0.2011 Acc: 0.9346
Epoch 16/24
----------
train Loss: 0.3006 Acc: 0.8770
val Loss: 0.1766 Acc: 0.9542
Epoch 17/24
----------
train Loss: 0.3397 Acc: 0.8443
val Loss: 0.2180 Acc: 0.9412
Epoch 18/24
----------
train Loss: 0.3332 Acc: 0.8443
val Loss: 0.1928 Acc: 0.9477
Epoch 19/24
----------
train Loss: 0.3563 Acc: 0.8238
val Loss: 0.1982 Acc: 0.9477
Epoch 20/24
----------
train Loss: 0.3222 Acc: 0.8566
val Loss: 0.2268 Acc: 0.9281
Epoch 21/24
----------
train Loss: 0.4554 Acc: 0.8115
val Loss: 0.2420 Acc: 0.9281
Epoch 22/24
----------
train Loss: 0.3066 Acc: 0.8648
val Loss: 0.1828 Acc: 0.9542
Epoch 23/24
----------
train Loss: 0.4099 Acc: 0.8279
val Loss: 0.2061 Acc: 0.9477
Epoch 24/24
----------
train Loss: 0.3176 Acc: 0.8648
val Loss: 0.2098 Acc: 0.9346
Training complete in 0m 34s
Best val Acc: 0.954248
（2）模型评测结果可视化
visualize_model(model_conv)
plt.ioff()
plt.show()
- 输出
