Transfer Learning for Computer Vision Tutorial

Author: Sasank Chilamkurthy

In this tutorial, you will learn how to train a convolutional neural network for image classification 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':
                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)
            if phase == 'train':
                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
            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
# Here the size of each output sample is set to 2.
# Alternatively, it can be generalized to nn.Linear(num_ftrs, len(class_names)).
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: 0.5864 Acc: 0.7008
val Loss: 0.3089 Acc: 0.8693

Epoch 1/24
----------
train Loss: 0.4400 Acc: 0.8033
val Loss: 0.4577 Acc: 0.8105

Epoch 2/24
----------
train Loss: 0.5539 Acc: 0.7705
val Loss: 0.2401 Acc: 0.9020

Epoch 3/24
----------
train Loss: 0.4871 Acc: 0.8115
val Loss: 1.3367 Acc: 0.6209

Epoch 4/24
----------
train Loss: 0.6163 Acc: 0.7377
val Loss: 0.3179 Acc: 0.8954

Epoch 5/24
----------
train Loss: 0.6211 Acc: 0.8156
val Loss: 0.3462 Acc: 0.8889

Epoch 6/24
----------
train Loss: 0.4831 Acc: 0.8197
val Loss: 0.5663 Acc: 0.8039

Epoch 7/24
----------
train Loss: 0.4639 Acc: 0.8115
val Loss: 0.2843 Acc: 0.9216

Epoch 8/24
----------
train Loss: 0.4221 Acc: 0.8484
val Loss: 0.2764 Acc: 0.9412

Epoch 9/24
----------
train Loss: 0.2547 Acc: 0.8893
val Loss: 0.2860 Acc: 0.9150

Epoch 10/24
----------
train Loss: 0.3601 Acc: 0.8648
val Loss: 0.3159 Acc: 0.8889

Epoch 11/24
----------
train Loss: 0.3852 Acc: 0.8402
val Loss: 0.3518 Acc: 0.8562

Epoch 12/24
----------
train Loss: 0.2783 Acc: 0.8811
val Loss: 0.2573 Acc: 0.9346

Epoch 13/24
----------
train Loss: 0.2747 Acc: 0.8811
val Loss: 0.2584 Acc: 0.9346

Epoch 14/24
----------
train Loss: 0.2907 Acc: 0.8811
val Loss: 0.2623 Acc: 0.9281

Epoch 15/24
----------
train Loss: 0.2568 Acc: 0.8893
val Loss: 0.2645 Acc: 0.9281

Epoch 16/24
----------
train Loss: 0.2741 Acc: 0.8689
val Loss: 0.2595 Acc: 0.9346

Epoch 17/24
----------
train Loss: 0.4116 Acc: 0.7705
val Loss: 0.2691 Acc: 0.9216

Epoch 18/24
----------
train Loss: 0.3170 Acc: 0.8607
val Loss: 0.2622 Acc: 0.9346

Epoch 19/24
----------
train Loss: 0.2712 Acc: 0.8975
val Loss: 0.2566 Acc: 0.9542

Epoch 20/24
----------
train Loss: 0.2896 Acc: 0.8730
val Loss: 0.2555 Acc: 0.9346

Epoch 21/24
----------
train Loss: 0.2115 Acc: 0.9262
val Loss: 0.2729 Acc: 0.9346

Epoch 22/24
----------
train Loss: 0.2550 Acc: 0.9016
val Loss: 0.2833 Acc: 0.8954

Epoch 23/24
----------
train Loss: 0.2637 Acc: 0.9016
val Loss: 0.2664 Acc: 0.9216

Epoch 24/24
----------
train Loss: 0.2073 Acc: 0.9057
val Loss: 0.2678 Acc: 0.9281

Training complete in 1m 7s
Best val Acc: 0.954248

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: 0.7797 Acc: 0.5656
val Loss: 0.2664 Acc: 0.8889

Epoch 1/24
----------
train Loss: 0.4624 Acc: 0.7910
val Loss: 0.3015 Acc: 0.8693

Epoch 2/24
----------
train Loss: 0.3703 Acc: 0.8074
val Loss: 0.1814 Acc: 0.9477

Epoch 3/24
----------
train Loss: 0.4092 Acc: 0.7992
val Loss: 0.1543 Acc: 0.9608

Epoch 4/24
----------
train Loss: 0.4118 Acc: 0.8197
val Loss: 0.2435 Acc: 0.9085

Epoch 5/24
----------
train Loss: 0.4231 Acc: 0.8484
val Loss: 0.1838 Acc: 0.9281

Epoch 6/24
----------
train Loss: 0.4498 Acc: 0.8115
val Loss: 0.1928 Acc: 0.9281

Epoch 7/24
----------
train Loss: 0.3473 Acc: 0.8484
val Loss: 0.2123 Acc: 0.9412

Epoch 8/24
----------
train Loss: 0.4186 Acc: 0.8279
val Loss: 0.1729 Acc: 0.9477

Epoch 9/24
----------
train Loss: 0.3993 Acc: 0.8279
val Loss: 0.2131 Acc: 0.9346

Epoch 10/24
----------
train Loss: 0.4425 Acc: 0.8156
val Loss: 0.1705 Acc: 0.9477

Epoch 11/24
----------
train Loss: 0.3184 Acc: 0.8607
val Loss: 0.1854 Acc: 0.9412

Epoch 12/24
----------
train Loss: 0.3320 Acc: 0.8402
val Loss: 0.1719 Acc: 0.9542

Epoch 13/24
----------
train Loss: 0.3401 Acc: 0.8484
val Loss: 0.1726 Acc: 0.9412

Epoch 14/24
----------
train Loss: 0.2857 Acc: 0.8770
val Loss: 0.1861 Acc: 0.9346

Epoch 15/24
----------
train Loss: 0.4092 Acc: 0.8074
val Loss: 0.1738 Acc: 0.9346

Epoch 16/24
----------
train Loss: 0.4654 Acc: 0.7869
val Loss: 0.1756 Acc: 0.9477

Epoch 17/24
----------
train Loss: 0.3246 Acc: 0.8320
val Loss: 0.1673 Acc: 0.9608

Epoch 18/24
----------
train Loss: 0.3156 Acc: 0.8689
val Loss: 0.1757 Acc: 0.9542

Epoch 19/24
----------
train Loss: 0.3252 Acc: 0.8484
val Loss: 0.1634 Acc: 0.9412

Epoch 20/24
----------
train Loss: 0.3029 Acc: 0.8484
val Loss: 0.2048 Acc: 0.9281

Epoch 21/24
----------
train Loss: 0.3615 Acc: 0.8525
val Loss: 0.1718 Acc: 0.9542

Epoch 22/24
----------
train Loss: 0.3707 Acc: 0.8238
val Loss: 0.1873 Acc: 0.9412

Epoch 23/24
----------
train Loss: 0.3267 Acc: 0.8689
val Loss: 0.2028 Acc: 0.9346

Epoch 24/24
----------
train Loss: 0.3243 Acc: 0.8320
val Loss: 0.1833 Acc: 0.9346

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

visualize_model(model_conv)

plt.ioff()
plt.show()

Further Learning

If you would like to learn more about the applications of transfer learning, checkout our Quantized Transfer Learning for Computer Vision Tutorial.