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 initialization, 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: 0.5014 Acc: 0.7418
val Loss: 0.5128 Acc: 0.7516

Epoch 1/24
----------
train Loss: 0.5473 Acc: 0.7623
val Loss: 0.3461 Acc: 0.8758

Epoch 2/24
----------
train Loss: 0.5948 Acc: 0.7582
val Loss: 0.2629 Acc: 0.8824

Epoch 3/24
----------
train Loss: 0.5550 Acc: 0.7910
val Loss: 0.1964 Acc: 0.9216

Epoch 4/24
----------
train Loss: 0.4094 Acc: 0.8238
val Loss: 0.1975 Acc: 0.9477

Epoch 5/24
----------
train Loss: 0.7609 Acc: 0.7377
val Loss: 0.1903 Acc: 0.9412

Epoch 6/24
----------
train Loss: 0.3257 Acc: 0.8811
val Loss: 0.2550 Acc: 0.9020

Epoch 7/24
----------
train Loss: 0.3687 Acc: 0.8402
val Loss: 0.2317 Acc: 0.9150

Epoch 8/24
----------
train Loss: 0.3523 Acc: 0.8770
val Loss: 0.2225 Acc: 0.9216

Epoch 9/24
----------
train Loss: 0.2461 Acc: 0.8893
val Loss: 0.2018 Acc: 0.9216

Epoch 10/24
----------
train Loss: 0.3259 Acc: 0.8566
val Loss: 0.2091 Acc: 0.9281

Epoch 11/24
----------
train Loss: 0.3121 Acc: 0.8689
val Loss: 0.1841 Acc: 0.9346

Epoch 12/24
----------
train Loss: 0.3015 Acc: 0.8811
val Loss: 0.2172 Acc: 0.9281

Epoch 13/24
----------
train Loss: 0.2824 Acc: 0.8811
val Loss: 0.1885 Acc: 0.9412

Epoch 14/24
----------
train Loss: 0.2368 Acc: 0.9057
val Loss: 0.1841 Acc: 0.9216

Epoch 15/24
----------
train Loss: 0.2567 Acc: 0.8852
val Loss: 0.1872 Acc: 0.9412

Epoch 16/24
----------
train Loss: 0.2497 Acc: 0.8975
val Loss: 0.2133 Acc: 0.9216

Epoch 17/24
----------
train Loss: 0.1552 Acc: 0.9262
val Loss: 0.1844 Acc: 0.9412

Epoch 18/24
----------
train Loss: 0.2848 Acc: 0.8607
val Loss: 0.1868 Acc: 0.9281

Epoch 19/24
----------
train Loss: 0.3129 Acc: 0.8648
val Loss: 0.1790 Acc: 0.9281

Epoch 20/24
----------
train Loss: 0.2879 Acc: 0.9057
val Loss: 0.1792 Acc: 0.9281

Epoch 21/24
----------
train Loss: 0.1964 Acc: 0.9098
val Loss: 0.1735 Acc: 0.9477

Epoch 22/24
----------
train Loss: 0.2784 Acc: 0.8730
val Loss: 0.1809 Acc: 0.9281

Epoch 23/24
----------
train Loss: 0.3156 Acc: 0.8770
val Loss: 0.1936 Acc: 0.9346

Epoch 24/24
----------
train Loss: 0.2130 Acc: 0.9221
val Loss: 0.1827 Acc: 0.9281

Training complete in 1m 8s
Best val Acc: 0.947712

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: 0.5784 Acc: 0.6680
val Loss: 0.2170 Acc: 0.9346

Epoch 1/24
----------
train Loss: 0.5237 Acc: 0.7787
val Loss: 0.2054 Acc: 0.9216

Epoch 2/24
----------
train Loss: 0.5363 Acc: 0.7623
val Loss: 0.1914 Acc: 0.9412

Epoch 3/24
----------
train Loss: 0.4304 Acc: 0.8320
val Loss: 0.3092 Acc: 0.8824

Epoch 4/24
----------
train Loss: 0.4789 Acc: 0.8074
val Loss: 0.2028 Acc: 0.9477

Epoch 5/24
----------
train Loss: 0.4668 Acc: 0.7869
val Loss: 0.2984 Acc: 0.8954

Epoch 6/24
----------
train Loss: 0.4395 Acc: 0.7992
val Loss: 0.1851 Acc: 0.9477

Epoch 7/24
----------
train Loss: 0.4193 Acc: 0.8484
val Loss: 0.1927 Acc: 0.9346

Epoch 8/24
----------
train Loss: 0.3840 Acc: 0.8033
val Loss: 0.1973 Acc: 0.9412

Epoch 9/24
----------
train Loss: 0.2388 Acc: 0.8934
val Loss: 0.1972 Acc: 0.9346

Epoch 10/24
----------
train Loss: 0.2875 Acc: 0.8893
val Loss: 0.1900 Acc: 0.9346

Epoch 11/24
----------
train Loss: 0.4052 Acc: 0.8238
val Loss: 0.1912 Acc: 0.9477

Epoch 12/24
----------
train Loss: 0.3382 Acc: 0.8484
val Loss: 0.2103 Acc: 0.9150

Epoch 13/24
----------
train Loss: 0.3427 Acc: 0.8484
val Loss: 0.1749 Acc: 0.9477

Epoch 14/24
----------
train Loss: 0.3267 Acc: 0.8484
val Loss: 0.2580 Acc: 0.9085

Epoch 15/24
----------
train Loss: 0.2980 Acc: 0.8852
val Loss: 0.1846 Acc: 0.9477

Epoch 16/24
----------
train Loss: 0.2793 Acc: 0.8770
val Loss: 0.1875 Acc: 0.9477

Epoch 17/24
----------
train Loss: 0.3443 Acc: 0.8566
val Loss: 0.2240 Acc: 0.9216

Epoch 18/24
----------
train Loss: 0.3463 Acc: 0.8443
val Loss: 0.2010 Acc: 0.9281

Epoch 19/24
----------
train Loss: 0.3084 Acc: 0.8443
val Loss: 0.2062 Acc: 0.9281

Epoch 20/24
----------
train Loss: 0.2792 Acc: 0.8525
val Loss: 0.1837 Acc: 0.9542

Epoch 21/24
----------
train Loss: 0.3389 Acc: 0.8607
val Loss: 0.2274 Acc: 0.9216

Epoch 22/24
----------
train Loss: 0.3878 Acc: 0.8443
val Loss: 0.1914 Acc: 0.9346

Epoch 23/24
----------
train Loss: 0.3123 Acc: 0.8689
val Loss: 0.2145 Acc: 0.9216

Epoch 24/24
----------
train Loss: 0.4084 Acc: 0.8033
val Loss: 0.1881 Acc: 0.9477

Training complete in 0m 34s
Best val Acc: 0.954248

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.