Transfer Learning Tutorial

Author: Sasank Chilamkurthy

In this tutorial, you will learn how to train your network using transfer learning. You can read more about the transfer learning at cs231n notes

Quoting these notes,

In practice, very few people train an entire Convolutional Network from scratch (with random initialization), because it is relatively rare to have a dataset of sufficient size. Instead, it is common to pretrain a ConvNet on a very large dataset (e.g. ImageNet, which contains 1.2 million images with 1000 categories), and then use the ConvNet either as an initialization or a fixed feature extractor for the task of interest.

These two major transfer learning scenarios look as follows:

• Finetuning the convnet: Instead of random initializaion, we initialize the network with a pretrained network, like the one that is trained on imagenet 1000 dataset. Rest of the training looks as usual.
• ConvNet as fixed feature extractor: Here, we will freeze the weights for all of the network except that of the final fully connected layer. This last fully connected layer is replaced with a new one with random weights and only this layer is trained.

# 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

Load Data

We will use torchvision and torch.utils.data packages for loading the data.

The problem we’re going to solve today is to train a model to classify ants and bees. We have about 120 training images each for ants and bees. There are 75 validation images for each class. Usually, this is a very small dataset to generalize upon, if trained from scratch. Since we are using transfer learning, we should be able to generalize reasonably well.

This dataset is a very small subset of imagenet.

Note

Download the data from here and extract it to the current directory.

# 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 = '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")

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."""
    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

Now, let’s write a general function to train a model. Here, we will illustrate:

• Scheduling the learning rate
• Saving the best model

In the following, parameter scheduler is an LR scheduler object from torch.optim.lr_scheduler.

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
                # 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()

                # 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

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('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.

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)

Train and evaluate

It should take around 15-25 min on CPU. On GPU though, it takes less than a minute.

model_ft = train_model(model_ft, criterion, optimizer_ft, exp_lr_scheduler,
                       num_epochs=25)

Out:

Epoch 0/24
----------
train Loss: 0.6714 Acc: 0.6189
val Loss: 0.3327 Acc: 0.8824

Epoch 1/24
----------
train Loss: 0.5290 Acc: 0.7459
val Loss: 0.3340 Acc: 0.8693

Epoch 2/24
----------
train Loss: 0.4808 Acc: 0.8115
val Loss: 0.4147 Acc: 0.8497

Epoch 3/24
----------
train Loss: 0.3993 Acc: 0.8566
val Loss: 0.2157 Acc: 0.9281

Epoch 4/24
----------
train Loss: 0.4375 Acc: 0.8361
val Loss: 0.2544 Acc: 0.8954

Epoch 5/24
----------
train Loss: 0.3968 Acc: 0.8238
val Loss: 0.3284 Acc: 0.8889

Epoch 6/24
----------
train Loss: 0.4003 Acc: 0.8361
val Loss: 0.1793 Acc: 0.9412

Epoch 7/24
----------
train Loss: 0.3302 Acc: 0.8689
val Loss: 0.1650 Acc: 0.9477

Epoch 8/24
----------
train Loss: 0.2979 Acc: 0.9016
val Loss: 0.1684 Acc: 0.9542

Epoch 9/24
----------
train Loss: 0.3382 Acc: 0.8402
val Loss: 0.2173 Acc: 0.9477

Epoch 10/24
----------
train Loss: 0.2700 Acc: 0.8893
val Loss: 0.1925 Acc: 0.9477

Epoch 11/24
----------
train Loss: 0.1764 Acc: 0.9344
val Loss: 0.1785 Acc: 0.9346

Epoch 12/24
----------
train Loss: 0.2761 Acc: 0.9098
val Loss: 0.1739 Acc: 0.9477

Epoch 13/24
----------
train Loss: 0.3115 Acc: 0.8607
val Loss: 0.1846 Acc: 0.9412

Epoch 14/24
----------
train Loss: 0.2409 Acc: 0.8975
val Loss: 0.1836 Acc: 0.9412

Epoch 15/24
----------
train Loss: 0.2339 Acc: 0.9098
val Loss: 0.1975 Acc: 0.9281

Epoch 16/24
----------
train Loss: 0.2637 Acc: 0.8852
val Loss: 0.1796 Acc: 0.9412

Epoch 17/24
----------
train Loss: 0.2430 Acc: 0.8852
val Loss: 0.1908 Acc: 0.9412

Epoch 18/24
----------
train Loss: 0.3220 Acc: 0.8811
val Loss: 0.1982 Acc: 0.9477

Epoch 19/24
----------
train Loss: 0.3218 Acc: 0.8484
val Loss: 0.1897 Acc: 0.9477

Epoch 20/24
----------
train Loss: 0.2660 Acc: 0.8566
val Loss: 0.1746 Acc: 0.9412

Epoch 21/24
----------
train Loss: 0.2781 Acc: 0.8689
val Loss: 0.1796 Acc: 0.9542

Epoch 22/24
----------
train Loss: 0.3112 Acc: 0.8566
val Loss: 0.1803 Acc: 0.9608

Epoch 23/24
----------
train Loss: 0.2639 Acc: 0.8730
val Loss: 0.1843 Acc: 0.9412

Epoch 24/24
----------
train Loss: 0.2337 Acc: 0.9139
val Loss: 0.1796 Acc: 0.9608

Training complete in 1m 8s
Best val Acc: 0.960784

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().

You can read more about this in the documentation here.

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)

Train and evaluate

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)

Out:

Epoch 0/24
----------
train Loss: 0.6068 Acc: 0.6680
val Loss: 0.8091 Acc: 0.5817

Epoch 1/24
----------
train Loss: 0.5010 Acc: 0.7705
val Loss: 0.1770 Acc: 0.9608

Epoch 2/24
----------
train Loss: 0.5639 Acc: 0.7459
val Loss: 0.2618 Acc: 0.9020

Epoch 3/24
----------
train Loss: 0.4166 Acc: 0.7828
val Loss: 0.2105 Acc: 0.9412

Epoch 4/24
----------
train Loss: 0.5187 Acc: 0.7787
val Loss: 0.1916 Acc: 0.9477

Epoch 5/24
----------
train Loss: 0.4656 Acc: 0.7746
val Loss: 0.1775 Acc: 0.9542

Epoch 6/24
----------
train Loss: 0.2922 Acc: 0.8893
val Loss: 0.1974 Acc: 0.9477

Epoch 7/24
----------
train Loss: 0.4127 Acc: 0.8279
val Loss: 0.1912 Acc: 0.9477

Epoch 8/24
----------
train Loss: 0.3030 Acc: 0.8648
val Loss: 0.2013 Acc: 0.9412

Epoch 9/24
----------
train Loss: 0.2744 Acc: 0.8770
val Loss: 0.1964 Acc: 0.9477

Epoch 10/24
----------
train Loss: 0.3811 Acc: 0.8402
val Loss: 0.1980 Acc: 0.9477

Epoch 11/24
----------
train Loss: 0.3949 Acc: 0.8279
val Loss: 0.1820 Acc: 0.9542

Epoch 12/24
----------
train Loss: 0.4495 Acc: 0.8033
val Loss: 0.1970 Acc: 0.9412

Epoch 13/24
----------
train Loss: 0.3321 Acc: 0.8648
val Loss: 0.2065 Acc: 0.9281

Epoch 14/24
----------
train Loss: 0.3677 Acc: 0.8443
val Loss: 0.1955 Acc: 0.9412

Epoch 15/24
----------
train Loss: 0.2591 Acc: 0.9016
val Loss: 0.2168 Acc: 0.9412

Epoch 16/24
----------
train Loss: 0.2849 Acc: 0.8893
val Loss: 0.1854 Acc: 0.9542

Epoch 17/24
----------
train Loss: 0.2608 Acc: 0.8934
val Loss: 0.1877 Acc: 0.9477

Epoch 18/24
----------
train Loss: 0.3253 Acc: 0.8607
val Loss: 0.1997 Acc: 0.9477

Epoch 19/24
----------
train Loss: 0.3512 Acc: 0.8607
val Loss: 0.2102 Acc: 0.9477

Epoch 20/24
----------
train Loss: 0.3020 Acc: 0.8648
val Loss: 0.1831 Acc: 0.9542

Epoch 21/24
----------
train Loss: 0.2790 Acc: 0.8525
val Loss: 0.1910 Acc: 0.9477

Epoch 22/24
----------
train Loss: 0.3746 Acc: 0.8402
val Loss: 0.1844 Acc: 0.9542

Epoch 23/24
----------
train Loss: 0.3622 Acc: 0.8320
val Loss: 0.1881 Acc: 0.9412

Epoch 24/24
----------
train Loss: 0.3478 Acc: 0.8770
val Loss: 0.1877 Acc: 0.9412

Training complete in 0m 35s
Best val Acc: 0.960784

visualize_model(model_conv)

plt.ioff()
plt.show()