Transfer Learning tutorial


In [2]:
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

Load Data


In [4]:
# Data augmentation and normalization for training
# Just normalization for validation
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 = '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")

Visualize a few images


In [5]:
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


# Get a batch of training data
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 the model

Scheduling the learning rate
Saving the best model

In [8]:
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':
                scheduler.step()
                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
                #tack 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()
                        
                #statistics
                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))
            
            # 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

Visualizing the model predictions


In [10]:
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)

Finetuning the convnet

Load a pretrained model and reset final fully connected layer.


In [11]:
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()

# Observe that all parameters are being optimized
optimizer_ft = optim.SGD(model_ft.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_ft, step_size=7, gamma=0.1)


Downloading: "https://download.pytorch.org/models/resnet18-5c106cde.pth" to C:\Users\Max/.torch\models\resnet18-5c106cde.pth
100.0%

Train and evaluate


In [12]:
model_ft = train_model(model_ft, criterion, optimizer_ft, exp_lr_scheduler,
                       num_epochs=25)


Epoch0/24
----------
train Loss: 0.6360 Acc: 0.6516
val Loss: 0.2425 Acc: 0.9020

Epoch1/24
----------
train Loss: 0.5098 Acc: 0.7951
val Loss: 0.2395 Acc: 0.9020

Epoch2/24
----------
train Loss: 0.4974 Acc: 0.7664
val Loss: 0.3065 Acc: 0.8889

Epoch3/24
----------
train Loss: 0.6932 Acc: 0.7418
val Loss: 0.6494 Acc: 0.8039

Epoch4/24
----------
train Loss: 0.4778 Acc: 0.8115
val Loss: 0.4587 Acc: 0.8431

Epoch5/24
----------
train Loss: 0.5479 Acc: 0.7787
val Loss: 0.7103 Acc: 0.8105

Epoch6/24
----------
train Loss: 0.4232 Acc: 0.8484
val Loss: 0.2453 Acc: 0.8889

Epoch7/24
----------
train Loss: 0.4817 Acc: 0.8402
val Loss: 0.2461 Acc: 0.8954

Epoch8/24
----------
train Loss: 0.3184 Acc: 0.8566
val Loss: 0.2357 Acc: 0.9150

Epoch9/24
----------
train Loss: 0.3326 Acc: 0.8770
val Loss: 0.2101 Acc: 0.9216

Epoch10/24
----------
train Loss: 0.2576 Acc: 0.8852
val Loss: 0.1881 Acc: 0.9216

Epoch11/24
----------
train Loss: 0.3632 Acc: 0.8443
val Loss: 0.1852 Acc: 0.9216

Epoch12/24
----------
train Loss: 0.2754 Acc: 0.8689
val Loss: 0.1925 Acc: 0.9216

Epoch13/24
----------
train Loss: 0.3217 Acc: 0.8730
val Loss: 0.2047 Acc: 0.9281

Epoch14/24
----------
train Loss: 0.3102 Acc: 0.8484
val Loss: 0.1984 Acc: 0.9150

Epoch15/24
----------
train Loss: 0.3382 Acc: 0.8484
val Loss: 0.2188 Acc: 0.9085

Epoch16/24
----------
train Loss: 0.3170 Acc: 0.8566
val Loss: 0.1942 Acc: 0.9216

Epoch17/24
----------
train Loss: 0.2253 Acc: 0.8975
val Loss: 0.2032 Acc: 0.9150

Epoch18/24
----------
train Loss: 0.3558 Acc: 0.8320
val Loss: 0.1987 Acc: 0.9150

Epoch19/24
----------
train Loss: 0.1741 Acc: 0.9344
val Loss: 0.2056 Acc: 0.9216

Epoch20/24
----------
train Loss: 0.2362 Acc: 0.8975
val Loss: 0.2037 Acc: 0.9085

Epoch21/24
----------
train Loss: 0.2467 Acc: 0.8852
val Loss: 0.2048 Acc: 0.9281

Epoch22/24
----------
train Loss: 0.3196 Acc: 0.8648
val Loss: 0.2066 Acc: 0.9150

Epoch23/24
----------
train Loss: 0.3341 Acc: 0.8197
val Loss: 0.1975 Acc: 0.9346

Epoch24/24
----------
train Loss: 0.3270 Acc: 0.8525
val Loss: 0.2136 Acc: 0.9020

Training complete in 44m 9s
Best val Acc: 0.934641

In [13]:
visualize_model(model_ft)


ConvNet as fixed feature extractor

Here, we need to freeze all the network except the final layer. We need to set requires_grad == False to freeze the parameters so that the gradients are not computed in backward().


In [14]:
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
# opoosed 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)

Train and evaluate


In [15]:
model_conv = train_model(model_conv, criterion, optimizer_conv,
                         exp_lr_scheduler, num_epochs=25)


Epoch0/24
----------
train Loss: 0.7234 Acc: 0.5984
val Loss: 0.2953 Acc: 0.8824

Epoch1/24
----------
train Loss: 0.5705 Acc: 0.7213
val Loss: 0.2025 Acc: 0.9477

Epoch2/24
----------
train Loss: 0.5528 Acc: 0.7664
val Loss: 0.1724 Acc: 0.9608

Epoch3/24
----------
train Loss: 0.4432 Acc: 0.8156
val Loss: 0.2555 Acc: 0.9020

Epoch4/24
----------
train Loss: 0.6102 Acc: 0.7459
val Loss: 0.2622 Acc: 0.9085

Epoch5/24
----------
train Loss: 0.4171 Acc: 0.8279
val Loss: 0.2538 Acc: 0.9150

Epoch6/24
----------
train Loss: 0.5245 Acc: 0.7500
val Loss: 0.2101 Acc: 0.9216

Epoch7/24
----------
train Loss: 0.2744 Acc: 0.8770
val Loss: 0.1687 Acc: 0.9477

Epoch8/24
----------
train Loss: 0.3163 Acc: 0.8689
val Loss: 0.1789 Acc: 0.9477

Epoch9/24
----------
train Loss: 0.4025 Acc: 0.8238
val Loss: 0.1488 Acc: 0.9477

Epoch10/24
----------
train Loss: 0.3734 Acc: 0.8525
val Loss: 0.1709 Acc: 0.9346

Epoch11/24
----------
train Loss: 0.3168 Acc: 0.8566
val Loss: 0.1570 Acc: 0.9477

Epoch12/24
----------
train Loss: 0.4342 Acc: 0.8074
val Loss: 0.1687 Acc: 0.9412

Epoch13/24
----------
train Loss: 0.2806 Acc: 0.8607
val Loss: 0.1649 Acc: 0.9281

Epoch14/24
----------
train Loss: 0.2764 Acc: 0.8648
val Loss: 0.1506 Acc: 0.9542

Epoch15/24
----------
train Loss: 0.4174 Acc: 0.7992
val Loss: 0.1555 Acc: 0.9477

Epoch16/24
----------
train Loss: 0.3274 Acc: 0.8607
val Loss: 0.1819 Acc: 0.9281

Epoch17/24
----------
train Loss: 0.3583 Acc: 0.8607
val Loss: 0.1502 Acc: 0.9477

Epoch18/24
----------
train Loss: 0.3407 Acc: 0.8361
val Loss: 0.1691 Acc: 0.9412

Epoch19/24
----------
train Loss: 0.3933 Acc: 0.8156
val Loss: 0.1481 Acc: 0.9477

Epoch20/24
----------
train Loss: 0.2942 Acc: 0.8852
val Loss: 0.2028 Acc: 0.9412

Epoch21/24
----------
train Loss: 0.3542 Acc: 0.8402
val Loss: 0.1757 Acc: 0.9412

Epoch22/24
----------
train Loss: 0.4346 Acc: 0.8074
val Loss: 0.1718 Acc: 0.9477

Epoch23/24
----------
train Loss: 0.2799 Acc: 0.9057
val Loss: 0.1665 Acc: 0.9477

Epoch24/24
----------
train Loss: 0.3096 Acc: 0.8730
val Loss: 0.1623 Acc: 0.9477

Training complete in 19m 57s
Best val Acc: 0.960784

In [16]:
visualize_model(model_conv)

plt.ioff()
plt.show()



In [ ]: