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: 0.5645 Acc: 0.7172
val Loss: 0.1935 Acc: 0.9216

Epoch 1/24
----------
train Loss: 0.5682 Acc: 0.7869
val Loss: 0.3477 Acc: 0.8562

Epoch 2/24
----------
train Loss: 0.4268 Acc: 0.7910
val Loss: 0.2366 Acc: 0.9346

Epoch 3/24
----------
train Loss: 0.4345 Acc: 0.8320
val Loss: 0.4671 Acc: 0.8497

Epoch 4/24
----------
train Loss: 0.5440 Acc: 0.7787
val Loss: 0.4988 Acc: 0.8170

Epoch 5/24
----------
train Loss: 0.6501 Acc: 0.7500
val Loss: 0.2597 Acc: 0.9346

Epoch 6/24
----------
train Loss: 0.4965 Acc: 0.7951
val Loss: 0.2806 Acc: 0.8889

Epoch 7/24
----------
train Loss: 0.4626 Acc: 0.7869
val Loss: 0.2783 Acc: 0.9085

Epoch 8/24
----------
train Loss: 0.2908 Acc: 0.8811
val Loss: 0.2571 Acc: 0.9150

Epoch 9/24
----------
train Loss: 0.3130 Acc: 0.8689
val Loss: 0.2318 Acc: 0.9150

Epoch 10/24
----------
train Loss: 0.2834 Acc: 0.8770
val Loss: 0.2513 Acc: 0.9281

Epoch 11/24
----------
train Loss: 0.3086 Acc: 0.8893
val Loss: 0.2317 Acc: 0.9281

Epoch 12/24
----------
train Loss: 0.3007 Acc: 0.8525
val Loss: 0.2510 Acc: 0.9085

Epoch 13/24
----------
train Loss: 0.2004 Acc: 0.9262
val Loss: 0.2205 Acc: 0.9150

Epoch 14/24
----------
train Loss: 0.2113 Acc: 0.9180
val Loss: 0.2413 Acc: 0.9281

Epoch 15/24
----------
train Loss: 0.2531 Acc: 0.8934
val Loss: 0.2323 Acc: 0.9150

Epoch 16/24
----------
train Loss: 0.2258 Acc: 0.8975
val Loss: 0.2383 Acc: 0.9281

Epoch 17/24
----------
train Loss: 0.2669 Acc: 0.8770
val Loss: 0.2569 Acc: 0.9281

Epoch 18/24
----------
train Loss: 0.2449 Acc: 0.8893
val Loss: 0.2427 Acc: 0.9216

Epoch 19/24
----------
train Loss: 0.2668 Acc: 0.8852
val Loss: 0.2229 Acc: 0.9150

Epoch 20/24
----------
train Loss: 0.3111 Acc: 0.8443
val Loss: 0.2163 Acc: 0.9150

Epoch 21/24
----------
train Loss: 0.2874 Acc: 0.8525
val Loss: 0.2203 Acc: 0.9150

Epoch 22/24
----------
train Loss: 0.2555 Acc: 0.9139
val Loss: 0.2378 Acc: 0.9281

Epoch 23/24
----------
train Loss: 0.2509 Acc: 0.8852
val Loss: 0.2174 Acc: 0.9150

Epoch 24/24
----------
train Loss: 0.3372 Acc: 0.8648
val Loss: 0.2108 Acc: 0.9150

Training complete in 1m 6s
Best val Acc: 0.934641

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: 0.5559 Acc: 0.6967
val Loss: 0.2345 Acc: 0.9412

Epoch 1/24
----------
train Loss: 0.4030 Acc: 0.8320
val Loss: 0.2549 Acc: 0.8954

Epoch 2/24
----------
train Loss: 0.4171 Acc: 0.8320
val Loss: 0.1855 Acc: 0.9542

Epoch 3/24
----------
train Loss: 0.5828 Acc: 0.7254
val Loss: 0.1929 Acc: 0.9542

Epoch 4/24
----------
train Loss: 0.5945 Acc: 0.7336
val Loss: 0.2259 Acc: 0.9281

Epoch 5/24
----------
train Loss: 0.5821 Acc: 0.7582
val Loss: 0.2925 Acc: 0.9085

Epoch 6/24
----------
train Loss: 0.4769 Acc: 0.7787
val Loss: 0.1802 Acc: 0.9608

Epoch 7/24
----------
train Loss: 0.3113 Acc: 0.8852
val Loss: 0.1809 Acc: 0.9542

Epoch 8/24
----------
train Loss: 0.3557 Acc: 0.8402
val Loss: 0.2530 Acc: 0.9281

Epoch 9/24
----------
train Loss: 0.2665 Acc: 0.8934
val Loss: 0.1968 Acc: 0.9608

Epoch 10/24
----------
train Loss: 0.3350 Acc: 0.8607
val Loss: 0.2173 Acc: 0.9346

Epoch 11/24
----------
train Loss: 0.3671 Acc: 0.8402
val Loss: 0.1931 Acc: 0.9477

Epoch 12/24
----------
train Loss: 0.3707 Acc: 0.8279
val Loss: 0.1961 Acc: 0.9477

Epoch 13/24
----------
train Loss: 0.3566 Acc: 0.8525
val Loss: 0.2017 Acc: 0.9477

Epoch 14/24
----------
train Loss: 0.3117 Acc: 0.8689
val Loss: 0.2088 Acc: 0.9477

Epoch 15/24
----------
train Loss: 0.4421 Acc: 0.7951
val Loss: 0.2445 Acc: 0.9281

Epoch 16/24
----------
train Loss: 0.3764 Acc: 0.8361
val Loss: 0.1977 Acc: 0.9477

Epoch 17/24
----------
train Loss: 0.4191 Acc: 0.8115
val Loss: 0.2160 Acc: 0.9346

Epoch 18/24
----------
train Loss: 0.3668 Acc: 0.8566
val Loss: 0.1916 Acc: 0.9608

Epoch 19/24
----------
train Loss: 0.3399 Acc: 0.8689
val Loss: 0.1976 Acc: 0.9412

Epoch 20/24
----------
train Loss: 0.3560 Acc: 0.8320
val Loss: 0.1844 Acc: 0.9608

Epoch 21/24
----------
train Loss: 0.2653 Acc: 0.8811
val Loss: 0.1948 Acc: 0.9542

Epoch 22/24
----------
train Loss: 0.3719 Acc: 0.8279
val Loss: 0.2132 Acc: 0.9412

Epoch 23/24
----------
train Loss: 0.4099 Acc: 0.8238
val Loss: 0.2215 Acc: 0.9281

Epoch 24/24
----------
train Loss: 0.3163 Acc: 0.8525
val Loss: 0.2100 Acc: 0.9477

Training complete in 0m 34s
Best val Acc: 0.960784

visualize_model(model_conv)

plt.ioff()
plt.show()