# 轉移學習的兩個主要場景：
# 1微調Convnet：使用預訓練的網絡(如在imagenet 1000上訓練而來的網絡)來初始化自己的網絡，
# 而不是隨機初始化。其他的訓練步驟不變。
# 2將Convnet看成固定的特徵提取器:首先固定ConvNet除了最後的全連接層外的其他所有層。最後的全連
# 接層被替換成一個新的隨機 初始化的層，只有這個新的層會被訓練[只有這層參數會在反向傳播時更新]
# 下面是利用PyTorch進行遷移學習步驟，要解決的問題是訓練一個模型來對螞蟻和蜜蜂進行分類。
# License: BSD
# Author: Sasank Chilamkurthy
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() # interactive mode
# 加載數據
# 今天要解決的問題是訓練一個模型來分類螞蟻ants和蜜蜂bees。ants和bees各有約120張訓練圖片。每個類
# 有75張驗證圖片。從零開始在 如此小的數據集上進行訓練通常是很難泛化的。由於我們使用遷移學習，模型
# 的泛化能力會相當好。 該數據集是imagenet的一個非常小的子集。從此處(https://download.pytorch.org
# /tutorial/hymenoptera_data.zip)下載數據，並將其解壓縮到當前目錄。
#訓練集數據擴充和歸一化
#在驗證集上僅需要歸一化
data_transforms = {
'train': transforms.Compose([
transforms.RandomResizedCrop(224), #隨機裁剪一個area然後再resize
transforms.RandomHorizontalFlip(), #隨機水平翻轉
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])
]),
} #ToTensor()能夠把灰度範圍從0-255變換到0-1之間,而後面的transform.Normalize()則把0-1變換到(-1,1)
#torchvision.transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
# [0.485, 0.456, 0.406]這一組平均值是從imagenet訓練集中抽樣算出來的。
# 對每個通道而言，Normalize執行以下操作：image=(image-mean)/std 如：(image-0.485)/0.229
# 不同數據集就有不同的標準化係數，例如([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])就是Imagenet dataset
# 的標準化係數（RGB三個通道對應三組係數），當需要將imagenet預訓練的參數遷移到另一神經網絡時，被遷移的神經網
# 絡就需要使用imagenet的係數，否則預訓練不僅無法起到應有的作用甚至還會幫倒忙，
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")
# 可視化部分圖像數據
# 可視化部分訓練圖像，以便了解數據擴充。
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
# 獲取一批訓練數據
inputs, classes = next(iter(dataloaders['train']))
# 批量製作網格
out = torchvision.utils.make_grid(inputs)
imshow(out, title=[class_names[x] for x in classes])
# 訓練模型
# 編寫一個通用函數來訓練模型。下面將說明： * 調整學習速率 * 保存最好的模型
# 下面的參數scheduler是一個來自 torch.optim.lr_scheduler的學習速率調整類的對象(LR scheduler 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)
# 每個epoch都有一個訓練和驗證階段
for phase in ['train', 'val']:
if phase == 'train':
scheduler.step()
model.train() # Set model to training mode
else:
model.eval() # Set model to evaluate mode
running_loss = 0.0
running_corrects = 0
# 迭代數據.
for inputs, labels in dataloaders[phase]:
inputs = inputs.to(device)
labels = labels.to(device)
# 零參數梯度
optimizer.zero_grad()
# 前向
# 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)
# 後向+僅在訓練階段進行優化
if phase == 'train':
loss.backward()
optimizer.step()
# 統計
running_loss += loss.item() * inputs.size(0)
running_corrects += torch.sum(preds == labels.data)
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))
# 深度複製mo
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))
# 加載最佳模型權重
model.load_state_dict(best_model_wts)
return model
# 可視化模型的預測結果
#一個通用的展示少量預測圖片的函數
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)
# 場景1：微調ConvNet
# 加載預訓練模型並重置最終完全連接的圖層。
model_ft = models.resnet18(pretrained=True)
num_ftrs = model_ft.fc.in_features
model_ft.fc = nn.Linear(num_ftrs, 2)
model_ft = model_ft.to(device)
criterion = nn.CrossEntropyLoss()
# 觀察所有參數都正在優化
optimizer_ft = optim.SGD(model_ft.parameters(), lr=0.001, momentum=0.9)
# 每7個epochs衰減LR通過設置gamma=0.1
exp_lr_scheduler = lr_scheduler.StepLR(optimizer_ft, step_size=7, gamma=0.1)
# 訓練和評估模型
# （1）訓練模型 該過程在CPU上需要大約15-25分鐘，但是在GPU上，它只需不到一分鐘。
model_ft = train_model(model_ft, criterion, optimizer_ft, exp_lr_scheduler,
num_epochs=25)
# 輸出:
Epoch 0/24
----------
train Loss: 0.7032 Acc: 0.6025
val Loss: 0.1698 Acc: 0.9412
Epoch 1/24
----------
train Loss: 0.6411 Acc: 0.7787
val Loss: 0.1981 Acc: 0.9281
·
·
·
Epoch 24/24
----------
train Loss: 0.2812 Acc: 0.8730
val Loss: 0.2647 Acc: 0.9150
Training complete in 1m 7s
Best val Acc: 0.941176
# （2）模型評估效果可視化
visualize_model(model_ft)
# 場景2：ConvNet作爲固定特徵提取器
# 在這裏需要凍結除最後一層之外的所有網絡。通過設置requires_grad == Falsebackward()來凍結參數，
# 這樣在反向傳播backward()的時候他們的梯度就不會被計算。
model_conv = torchvision.models.resnet18(pretrained=True)
for param in model_conv.parameters():
param.requires_grad = False
# Parameters of newly constructed modules have requires_grad=True by default
num_ftrs = model_conv.fc.in_features
model_conv.fc = nn.Linear(num_ftrs, 2)
model_conv = model_conv.to(device)
criterion = nn.CrossEntropyLoss()
# Observe that only parameters of final layer are being optimized as
# opposed to before.
optimizer_conv = optim.SGD(model_conv.fc.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_conv, step_size=7, gamma=0.1)
# 訓練和評估
# （1）訓練模型 在CPU上，與前一個場景相比，這將花費大約一半的時間，因爲不需要爲大多數網絡計算梯度。
# 但需要計算轉發
model_conv = train_model(model_conv, criterion, optimizer_conv,
exp_lr_scheduler, num_epochs=25)
# 輸出
Epoch 0/24
----------
train Loss: 0.6400 Acc: 0.6434
val Loss: 0.2539 Acc: 0.9085
·
·
·
Epoch 23/24
----------
train Loss: 0.2988 Acc: 0.8607
val Loss: 0.2151 Acc: 0.9412
Epoch 24/24
----------
train Loss: 0.3519 Acc: 0.8484
val Loss: 0.2045 Acc: 0.9412
Training complete in 0m 35s
Best val Acc: 0.954248
（2）模型評估效果可視化
visualize_model(model_conv)
plt.ioff()
plt.show()
