PyTorch 計算機視覺的遷移學習教程代碼詳解 （TRANSFER LEARNING FOR COMPUTER VISION TUTORIAL ）

PyTorch 原文：

https://pytorch.org/tutorials/beginner/transfer_learning_tutorial.html

參考文章：

https://www.cnblogs.com/king-lps/p/8665344.html

https://blog.csdn.net/shaopeng568/article/details/95205345

https://blog.csdn.net/yuyangyg/article/details/80018574

# 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 mod 打開交互模式e # Data augmentation and normalization for training # Just normalization for validation # torchvision.transforms模塊提供了一般的圖像轉換操作類，一般使用Compose把多個步驟整合到一起 data_transforms = { 'train': transforms.Compose([ transforms.RandomResizedCrop(224), # 先將給定的PIL.Image隨機切，然后再resize成給定的尺寸大小,裁剪到224*224 transforms.RandomHorizontalFlip(), # 隨機水平翻轉給定的PIL.Image，概率為0.5，即一半的概率翻轉，一半的概率不翻轉 transforms.ToTensor(), # 將 PIL Image 或者 ndarray 轉換為 tensor，並且歸一化至[0-1], shape=chw transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]) # 給定均值(R,G,B)，方差(R,G,B)，將會把tensor正則化，即 Normalized_image = (image-mean)/std ]), 'val': transforms.Compose([ transforms.Resize(256), # 將輸入的PIL圖像調整為給定的大小 transforms.CenterCrop(224), # 將給定的PIL.Image進行中心切割 transforms.ToTensor(), transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225]) # 對數據按通道進行標准化 ]), } # Tensor總結 # （1）Tensor 和 Numpy都是矩陣，區別是前者可以在GPU上運行，后者只能在CPU上； # （2）Tensor和Numpy互相轉化很方便，類型也比較兼容 # （3）Tensor可以直接通過print顯示數據類型，而Numpy不可以 # Visualize a few images # Let’s visualize a few training images so as to understand the data augmentations. def imshow(inp, title=None): """Imshow for Tensor.""" # transpose是求轉置矩陣函數 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 # clip(a, a_min, a_max, out=None)表示數組a中所有的數限定到范圍a_min和a_max中 inp = np.clip(inp, 0, 1) plt.imshow(inp) if title is not None: plt.title(title) plt.pause(1) # pause a bit so that plots are updated data_dir = 'hymenoptera_data' # ImageFolder的第一個參數是在指定的路徑下尋找圖片，第二個參數transform是指對圖像進行轉換操作 image_datasets = {x: datasets.ImageFolder(os.path.join(data_dir, x), data_transforms[x]) for x in ['train', 'val']} # dataLoader第一個參數是傳入的數據集，第二個參數batch_size表示每個batch有多少個樣本，shuffle表示在每個epoch開始的時候，對數據進行重新排序 # num_workers(int, optional)這個參數決定了有幾個進程來處理data loading。0意味着所有的數據都會被load進主進程（默認為0）。 # ！！！！！！！！！注意，這里num_workers的參數要改成0，不然在我電腦上就會出錯 dataloaders = {x: torch.utils.data.DataLoader(image_datasets[x], batch_size=4, shuffle=True, num_workers=0) for x in ['train', 'val']} # 得到訓練集和測試集的長度 dataset_sizes = {x: len(image_datasets[x]) for x in ['train', 'val']} # 得到類別的名稱 class_names = image_datasets['train'].classes # 檢驗是否可用cuda device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu") # ------------------- isualize a few images --------------------- # # Get a batch of training data inputs, classes = next(iter(dataloaders['train'])) # Make a grid from batch # mark_grid的作用是將若干幅圖像拼成一幅圖像。其中padding的作用就是子圖像與子圖像的pad有多寬 # 此時make_grid的輸入仍為Tensor(chw)，而imshow的時候要轉回hwc，而后要乘以方差並加上均值 out = torchvision.utils.make_grid(inputs) # imshow(out, title=[class_names[x] for x in classes]) # 導入預訓練的模型，如果pretrained=False或者為不帶參數表示只導入網絡結構，不導入參數 # model_ft即為含訓練好參數的殘差網絡 model_ft = models.resnet18(pretrained=True) # 最后一個全連接的輸入維度，這里實為512 num_ftrs = model_ft.fc.in_features # Here the size of each output sample is set to 2. # Alternatively, it can be generalized to nn.Linear(num_ftrs, len(class_names)). # 將最后一個全連接由（512， 1000）改成（512， 2） model_ft.fc = nn.Linear(num_ftrs, 2) # 將所有最開始讀取數據時的tensor變量copy一份到device所指定的GPU上去，之后的運算都在GPU上進行 model_ft = model_ft.to(device) # 定義計算損失的函數 criterion = nn.CrossEntropyLoss() # Observe that all parameters are being optimized # 神經網絡優化器，優化我們的網絡，使網絡在訓練過程中快起來，節省網絡訓練時間 # 要想優化網絡，並且先構造一個優化器 # SGD(stochastic Gradient Descent),SGD會把數據拆分后再分批不斷放入NN中計算，每次使用一批數據，雖然不能反映整體數據的情況，不過卻很 # 程度上加速了NN的訓練過程，而且不會跌勢太多准確率 # Momentum 傳統的參數 W 的更新是把原始的 W 累加上一個負的學習率(learning rate) 乘以校正值 (dx)。 # 我們把這個人從平地上放到了一個斜坡上, 只要他往下坡的方向走一點點, 由於向下的慣性, 他不自覺地就一直往下走, 走的彎路也變少了. 這就是 Momentum 參數更新 optimizer_ft = optim.SGD(model_ft.parameters(), lr=0.001, momentum=0.9) # Decay LR by a factor of 0.1 every 7 epochs # lr_scheduler是所有學習率改變策略的基類 # lr_scheduler提供了基於多種epoch數目調整學習率的方法 # step_size為int型，表示學習衰減期，指幾個epoch衰減一次，gamma為學習衰減的乘積因子 # 每7個epoch衰減0.1倍 exp_lr_scheduler = lr_scheduler.StepLR(optimizer_ft, step_size=7, gamma=0.1) # 定義樣本跑的次數 num_epochs = 25 # ----------------------- 訓練模型 ---------------------- # def train_model(model, criterion, optimizer, scheduler, num_epochs=25): # 計算模型初始開始訓練時間 since = time.time() # 先拷貝一份當前模型的參數，后面迭代過程中若遇到更優模型則替換 best_model_wts = copy.deepcopy(model.state_dict()) # 初始准確率 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，設置為訓練模式 else: model.eval() # Set model to evaluate mode，設置為測試模式 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 # 把梯度置零，把loss關於weight的倒數變成0 # 為下一次的訓練清空上一步的殘余更新參數值 optimizer.zero_grad() # forward # track history if only in train # set_grad_enabled用於設置梯度打開或關閉的上下文管理器 # set_grad_enabled將基於它的參數mode使用或禁用梯度。標記是否使能梯度，主要用在有條件的使能梯度 with torch.set_grad_enabled(phase == 'train'): # 前向傳播求出預測的值 outputs = model(inputs) _, preds = torch.max(outputs, 1) # 求損失 loss = criterion(outputs, labels) # backward + optimize only if in training phase if phase == 'train': # 誤差反向傳播，計算參數更新值 loss.backward() # 將參數更新值施加到網絡的參數上 optimizer.step() # statistics # 計算損失值和精確度 running_loss += loss.item() * inputs.size(0) running_corrects += torch.sum(preds == labels.data) if phase == 'train': # 應用概率學習率的策略，更新optimizer對象每個para_group字典的lr鍵的值，param_group['lr']=lr # 在訓練的時候進行迭代 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 # 當驗證時遇到了更好的模型則予以保留 if phase == 'val' and epoch_acc > best_acc: best_acc = epoch_acc # 拷貝模型參數 best_model_wts = copy.deepcopy(model.state_dict()) print() # 顯示計算模型訓練時間 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 # 載入最優模型參數 model.load_state_dict(best_model_wts) return model # model_ft = train_model(model_ft, criterion, optimizer_ft, exp_lr_scheduler, num_epochs=25) # ----------------- visualizing the model predictions ----------------- # def visualize_model(model, num_images=6): # 檢驗是否為訓練模型 was_training = model.training # 把模式設置為測試模式 model.eval() images_so_far = 0 fig = plt.figure() # operations inside don't track history # 使下面的計算圖不占用內存，不保存梯度，減小內存的占用 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) visualize_model(model_ft) # ----------------------------------- 局部微調 --------------------------------------- # # ConvNet as fixed feature extractor # 凍結除全連接層以外的最后所有層參數。我們需要設置 requires_grad == False 來凍結參數以便在反向過程中不需要計算梯度 model_conv = torchvision.models.resnet18(pretrained=True) for param in model_conv.parameters(): # 將所有參數求導設為否 param.requires_grad = False # Parameters of newly constructed modules have requires_grad=True by default # 得到最后一個全連接的輸入維度 num_ftrs = model_conv.fc.in_features # 將最后一個全連接由（512， 1000）改成（512， 2）,取代最后一個全連接 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. # ！！！！！！！注意，這里只有全連接層參數被優化 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) # On CPU this will take about half the time compared to previous scenario. # This is expected as gradients don’t need to be computed for most of the network. However, forward does need to be computed. model_conv = train_model(model_conv, criterion, optimizer_conv, exp_lr_scheduler, num_epochs=25) visualize_model(model_conv) # 顯示前關掉交叉模式 plt.ioff() plt.show() pass

本人初學水平，主要參考他人代碼，有錯誤歡迎指正