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: 0.5792 Acc: 0.7172
val Loss: 0.2619 Acc: 0.8889

Epoch 1/24
----------
train Loss: 0.3614 Acc: 0.8525
val Loss: 0.3318 Acc: 0.8889

Epoch 2/24
----------
train Loss: 0.4039 Acc: 0.8156
val Loss: 0.1952 Acc: 0.9216

Epoch 3/24
----------
train Loss: 0.6236 Acc: 0.7500
val Loss: 0.9116 Acc: 0.7582

Epoch 4/24
----------
train Loss: 0.6834 Acc: 0.7705
val Loss: 0.6322 Acc: 0.7778

Epoch 5/24
----------
train Loss: 0.5969 Acc: 0.7992
val Loss: 0.4023 Acc: 0.8431

Epoch 6/24
----------
train Loss: 0.5431 Acc: 0.7541
val Loss: 0.4407 Acc: 0.8105

Epoch 7/24
----------
train Loss: 0.3520 Acc: 0.8443
val Loss: 0.2905 Acc: 0.8562

Epoch 8/24
----------
train Loss: 0.3068 Acc: 0.8730
val Loss: 0.2869 Acc: 0.8758

Epoch 9/24
----------
train Loss: 0.3231 Acc: 0.8484
val Loss: 0.2684 Acc: 0.8954

Epoch 10/24
----------
train Loss: 0.3829 Acc: 0.8197
val Loss: 0.2472 Acc: 0.9085

Epoch 11/24
----------
train Loss: 0.3017 Acc: 0.8607
val Loss: 0.2321 Acc: 0.9150

Epoch 12/24
----------
train Loss: 0.4007 Acc: 0.8074
val Loss: 0.2384 Acc: 0.9085

Epoch 13/24
----------
train Loss: 0.2331 Acc: 0.9016
val Loss: 0.2795 Acc: 0.8889

Epoch 14/24
----------
train Loss: 0.2886 Acc: 0.8525
val Loss: 0.3140 Acc: 0.8693

Epoch 15/24
----------
train Loss: 0.2067 Acc: 0.9262
val Loss: 0.2609 Acc: 0.8889

Epoch 16/24
----------
train Loss: 0.2681 Acc: 0.8975
val Loss: 0.2376 Acc: 0.9020

Epoch 17/24
----------
train Loss: 0.2365 Acc: 0.8934
val Loss: 0.2624 Acc: 0.8889

Epoch 18/24
----------
train Loss: 0.2843 Acc: 0.9016
val Loss: 0.2471 Acc: 0.9020

Epoch 19/24
----------
train Loss: 0.2918 Acc: 0.8730
val Loss: 0.2576 Acc: 0.8889

Epoch 20/24
----------
train Loss: 0.2849 Acc: 0.8730
val Loss: 0.2791 Acc: 0.8889

Epoch 21/24
----------
train Loss: 0.3566 Acc: 0.8402
val Loss: 0.2108 Acc: 0.9281

Epoch 22/24
----------
train Loss: 0.2719 Acc: 0.8730
val Loss: 0.2209 Acc: 0.9216

Epoch 23/24
----------
train Loss: 0.2754 Acc: 0.8975
val Loss: 0.2281 Acc: 0.9150

Epoch 24/24
----------
train Loss: 0.2564 Acc: 0.8770
val Loss: 0.2255 Acc: 0.9216

Training complete in 1m 7s
Best val Acc: 0.928105

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: 0.6345 Acc: 0.6066
val Loss: 0.2096 Acc: 0.9412

Epoch 1/24
----------
train Loss: 0.4681 Acc: 0.7582
val Loss: 0.1720 Acc: 0.9477

Epoch 2/24
----------
train Loss: 0.5675 Acc: 0.7746
val Loss: 0.3037 Acc: 0.8758

Epoch 3/24
----------
train Loss: 0.4845 Acc: 0.8074
val Loss: 0.1916 Acc: 0.9477

Epoch 4/24
----------
train Loss: 0.5612 Acc: 0.7828
val Loss: 0.1858 Acc: 0.9477

Epoch 5/24
----------
train Loss: 0.4794 Acc: 0.8033
val Loss: 0.2075 Acc: 0.9346

Epoch 6/24
----------
train Loss: 0.5154 Acc: 0.7746
val Loss: 0.1944 Acc: 0.9477

Epoch 7/24
----------
train Loss: 0.3404 Acc: 0.8607
val Loss: 0.1785 Acc: 0.9542

Epoch 8/24
----------
train Loss: 0.2989 Acc: 0.8852
val Loss: 0.1841 Acc: 0.9542

Epoch 9/24
----------
train Loss: 0.2963 Acc: 0.8689
val Loss: 0.2010 Acc: 0.9412

Epoch 10/24
----------
train Loss: 0.4531 Acc: 0.7992
val Loss: 0.2034 Acc: 0.9477

Epoch 11/24
----------
train Loss: 0.3220 Acc: 0.8648
val Loss: 0.2095 Acc: 0.9281

Epoch 12/24
----------
train Loss: 0.3433 Acc: 0.8320
val Loss: 0.2219 Acc: 0.9281

Epoch 13/24
----------
train Loss: 0.3188 Acc: 0.8525
val Loss: 0.1875 Acc: 0.9608

Epoch 14/24
----------
train Loss: 0.3535 Acc: 0.8402
val Loss: 0.2170 Acc: 0.9412

Epoch 15/24
----------
train Loss: 0.3532 Acc: 0.8402
val Loss: 0.1963 Acc: 0.9542

Epoch 16/24
----------
train Loss: 0.2760 Acc: 0.8811
val Loss: 0.1867 Acc: 0.9412

Epoch 17/24
----------
train Loss: 0.3939 Acc: 0.8279
val Loss: 0.1988 Acc: 0.9542

Epoch 18/24
----------
train Loss: 0.3567 Acc: 0.8566
val Loss: 0.1880 Acc: 0.9542

Epoch 19/24
----------
train Loss: 0.3077 Acc: 0.8852
val Loss: 0.2245 Acc: 0.9346

Epoch 20/24
----------
train Loss: 0.3334 Acc: 0.8525
val Loss: 0.1897 Acc: 0.9477

Epoch 21/24
----------
train Loss: 0.3516 Acc: 0.8238
val Loss: 0.1883 Acc: 0.9542

Epoch 22/24
----------
train Loss: 0.2569 Acc: 0.8689
val Loss: 0.2112 Acc: 0.9412

Epoch 23/24
----------
train Loss: 0.3023 Acc: 0.8566
val Loss: 0.1993 Acc: 0.9542

Epoch 24/24
----------
train Loss: 0.3623 Acc: 0.8525
val Loss: 0.1978 Acc: 0.9412

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

visualize_model(model_conv)

plt.ioff()
plt.show()