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: 0.6095 Acc: 0.6475
val Loss: 0.2178 Acc: 0.9085

Epoch 1/24
----------
train Loss: 0.7108 Acc: 0.7049
val Loss: 0.2875 Acc: 0.8824

Epoch 2/24
----------
train Loss: 0.6654 Acc: 0.7049
val Loss: 0.3613 Acc: 0.8693

Epoch 3/24
----------
train Loss: 0.5540 Acc: 0.7910
val Loss: 0.2346 Acc: 0.9281

Epoch 4/24
----------
train Loss: 0.5656 Acc: 0.7828
val Loss: 0.4141 Acc: 0.9020

Epoch 5/24
----------
train Loss: 0.3808 Acc: 0.8320
val Loss: 0.4777 Acc: 0.8758

Epoch 6/24
----------
train Loss: 0.6411 Acc: 0.7869
val Loss: 0.3331 Acc: 0.8954

Epoch 7/24
----------
train Loss: 0.3745 Acc: 0.8730
val Loss: 0.2942 Acc: 0.9216

Epoch 8/24
----------
train Loss: 0.2934 Acc: 0.8975
val Loss: 0.2365 Acc: 0.9346

Epoch 9/24
----------
train Loss: 0.3276 Acc: 0.8525
val Loss: 0.2495 Acc: 0.9216

Epoch 10/24
----------
train Loss: 0.3791 Acc: 0.8361
val Loss: 0.2370 Acc: 0.9281

Epoch 11/24
----------
train Loss: 0.2845 Acc: 0.8607
val Loss: 0.2645 Acc: 0.9281

Epoch 12/24
----------
train Loss: 0.2829 Acc: 0.8730
val Loss: 0.2774 Acc: 0.9150

Epoch 13/24
----------
train Loss: 0.2755 Acc: 0.8852
val Loss: 0.2443 Acc: 0.9346

Epoch 14/24
----------
train Loss: 0.3581 Acc: 0.8197
val Loss: 0.2373 Acc: 0.9412

Epoch 15/24
----------
train Loss: 0.3116 Acc: 0.8648
val Loss: 0.2497 Acc: 0.9216

Epoch 16/24
----------
train Loss: 0.2347 Acc: 0.8975
val Loss: 0.2481 Acc: 0.9281

Epoch 17/24
----------
train Loss: 0.3021 Acc: 0.8893
val Loss: 0.2349 Acc: 0.9281

Epoch 18/24
----------
train Loss: 0.3420 Acc: 0.8320
val Loss: 0.2416 Acc: 0.9216

Epoch 19/24
----------
train Loss: 0.3057 Acc: 0.8934
val Loss: 0.2298 Acc: 0.9412

Epoch 20/24
----------
train Loss: 0.3714 Acc: 0.8443
val Loss: 0.2338 Acc: 0.9477

Epoch 21/24
----------
train Loss: 0.3370 Acc: 0.8484
val Loss: 0.2298 Acc: 0.9412

Epoch 22/24
----------
train Loss: 0.1977 Acc: 0.9221
val Loss: 0.2216 Acc: 0.9412

Epoch 23/24
----------
train Loss: 0.3429 Acc: 0.8566
val Loss: 0.2390 Acc: 0.9346

Epoch 24/24
----------
train Loss: 0.2376 Acc: 0.8893
val Loss: 0.2199 Acc: 0.9412

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: 0.5987 Acc: 0.6393
val Loss: 0.3066 Acc: 0.8693

Epoch 1/24
----------
train Loss: 0.5367 Acc: 0.7172
val Loss: 0.1962 Acc: 0.9477

Epoch 2/24
----------
train Loss: 0.4591 Acc: 0.7992
val Loss: 0.1974 Acc: 0.9477

Epoch 3/24
----------
train Loss: 0.5040 Acc: 0.7787
val Loss: 0.1922 Acc: 0.9346

Epoch 4/24
----------
train Loss: 0.5343 Acc: 0.7705
val Loss: 0.2062 Acc: 0.9542

Epoch 5/24
----------
train Loss: 0.5048 Acc: 0.7828
val Loss: 0.2255 Acc: 0.9281

Epoch 6/24
----------
train Loss: 0.5591 Acc: 0.7951
val Loss: 0.2150 Acc: 0.9346

Epoch 7/24
----------
train Loss: 0.3710 Acc: 0.8402
val Loss: 0.2361 Acc: 0.9346

Epoch 8/24
----------
train Loss: 0.2645 Acc: 0.8934
val Loss: 0.2024 Acc: 0.9346

Epoch 9/24
----------
train Loss: 0.3999 Acc: 0.8402
val Loss: 0.1959 Acc: 0.9477

Epoch 10/24
----------
train Loss: 0.3806 Acc: 0.8238
val Loss: 0.2191 Acc: 0.9346

Epoch 11/24
----------
train Loss: 0.4044 Acc: 0.8402
val Loss: 0.1941 Acc: 0.9542

Epoch 12/24
----------
train Loss: 0.3234 Acc: 0.8648
val Loss: 0.1977 Acc: 0.9477

Epoch 13/24
----------
train Loss: 0.3640 Acc: 0.8361
val Loss: 0.2026 Acc: 0.9346

Epoch 14/24
----------
train Loss: 0.4070 Acc: 0.8115
val Loss: 0.1912 Acc: 0.9542

Epoch 15/24
----------
train Loss: 0.3331 Acc: 0.8484
val Loss: 0.2011 Acc: 0.9346

Epoch 16/24
----------
train Loss: 0.3006 Acc: 0.8770
val Loss: 0.1766 Acc: 0.9542

Epoch 17/24
----------
train Loss: 0.3397 Acc: 0.8443
val Loss: 0.2180 Acc: 0.9412

Epoch 18/24
----------
train Loss: 0.3332 Acc: 0.8443
val Loss: 0.1928 Acc: 0.9477

Epoch 19/24
----------
train Loss: 0.3563 Acc: 0.8238
val Loss: 0.1982 Acc: 0.9477

Epoch 20/24
----------
train Loss: 0.3222 Acc: 0.8566
val Loss: 0.2268 Acc: 0.9281

Epoch 21/24
----------
train Loss: 0.4554 Acc: 0.8115
val Loss: 0.2420 Acc: 0.9281

Epoch 22/24
----------
train Loss: 0.3066 Acc: 0.8648
val Loss: 0.1828 Acc: 0.9542

Epoch 23/24
----------
train Loss: 0.4099 Acc: 0.8279
val Loss: 0.2061 Acc: 0.9477

Epoch 24/24
----------
train Loss: 0.3176 Acc: 0.8648
val Loss: 0.2098 Acc: 0.9346

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.