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

import torch
import torch.nn as nn
import torch.optim as optim
from torch.optim import lr_scheduler
import torch.backends.cudnn as cudnn
import numpy as np
import torchvision
from torchvision import datasets, models, transforms
import matplotlib.pyplot as plt
import time
import os
from PIL import Image
from tempfile import TemporaryDirectory

cudnn.benchmark = True
plt.ion()   # interactive mode

<contextlib.ExitStack object at 0x7f22b1e267a0>

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):
    """Display image 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()

    # Create a temporary directory to save training checkpoints
    with TemporaryDirectory() as tempdir:
        best_model_params_path = os.path.join(tempdir, 'best_model_params.pt')

        torch.save(model.state_dict(), best_model_params_path)
        best_acc = 0.0

        for epoch in range(num_epochs):
            print(f'Epoch {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(f'{phase} Loss: {epoch_loss:.4f} Acc: {epoch_acc:.4f}')

                # deep copy the model
                if phase == 'val' and epoch_acc > best_acc:
                    best_acc = epoch_acc
                    torch.save(model.state_dict(), best_model_params_path)

            print()

        time_elapsed = time.time() - since
        print(f'Training complete in {time_elapsed // 60:.0f}m {time_elapsed % 60:.0f}s')
        print(f'Best val Acc: {best_acc:4f}')

        # load best model weights
        model.load_state_dict(torch.load(best_model_params_path, weights_only=True))
    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(f'predicted: {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(weights='IMAGENET1K_V1')
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)

Downloading: "https://download.pytorch.org/models/resnet18-f37072fd.pth" to /var/lib/ci-user/.cache/torch/hub/checkpoints/resnet18-f37072fd.pth

  0%|          | 0.00/44.7M [00:00<?, ?B/s]
 46%|####6     | 20.6M/44.7M [00:00<00:00, 215MB/s]
 93%|#########3| 41.8M/44.7M [00:00<00:00, 219MB/s]
100%|##########| 44.7M/44.7M [00:00<00:00, 218MB/s]

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)

Epoch 0/24
----------
train Loss: 0.4768 Acc: 0.7623
val Loss: 0.3022 Acc: 0.8693

Epoch 1/24
----------
train Loss: 0.5255 Acc: 0.8156
val Loss: 0.6014 Acc: 0.7451

Epoch 2/24
----------
train Loss: 0.4475 Acc: 0.8361
val Loss: 0.2400 Acc: 0.9216

Epoch 3/24
----------
train Loss: 0.5749 Acc: 0.7787
val Loss: 0.3274 Acc: 0.8954

Epoch 4/24
----------
train Loss: 0.4023 Acc: 0.8607
val Loss: 0.2732 Acc: 0.9085

Epoch 5/24
----------
train Loss: 0.4900 Acc: 0.7828
val Loss: 0.2885 Acc: 0.8954

Epoch 6/24
----------
train Loss: 0.3938 Acc: 0.8074
val Loss: 0.4381 Acc: 0.8235

Epoch 7/24
----------
train Loss: 0.4091 Acc: 0.8074
val Loss: 0.1899 Acc: 0.9216

Epoch 8/24
----------
train Loss: 0.2178 Acc: 0.8934
val Loss: 0.1905 Acc: 0.9346

Epoch 9/24
----------
train Loss: 0.2704 Acc: 0.8852
val Loss: 0.1845 Acc: 0.9346

Epoch 10/24
----------
train Loss: 0.3568 Acc: 0.8648
val Loss: 0.1700 Acc: 0.9346

Epoch 11/24
----------
train Loss: 0.3156 Acc: 0.8443
val Loss: 0.2581 Acc: 0.8889

Epoch 12/24
----------
train Loss: 0.2264 Acc: 0.9098
val Loss: 0.2038 Acc: 0.9216

Epoch 13/24
----------
train Loss: 0.2889 Acc: 0.8852
val Loss: 0.1846 Acc: 0.9281

Epoch 14/24
----------
train Loss: 0.2893 Acc: 0.8893
val Loss: 0.2370 Acc: 0.9085

Epoch 15/24
----------
train Loss: 0.2970 Acc: 0.8811
val Loss: 0.2640 Acc: 0.9020

Epoch 16/24
----------
train Loss: 0.2035 Acc: 0.9098
val Loss: 0.1960 Acc: 0.9150

Epoch 17/24
----------
train Loss: 0.2452 Acc: 0.9057
val Loss: 0.1764 Acc: 0.9412

Epoch 18/24
----------
train Loss: 0.2817 Acc: 0.8811
val Loss: 0.2125 Acc: 0.9085

Epoch 19/24
----------
train Loss: 0.2074 Acc: 0.9180
val Loss: 0.1877 Acc: 0.9346

Epoch 20/24
----------
train Loss: 0.2479 Acc: 0.8934
val Loss: 0.2170 Acc: 0.9150

Epoch 21/24
----------
train Loss: 0.2463 Acc: 0.8852
val Loss: 0.2217 Acc: 0.9085

Epoch 22/24
----------
train Loss: 0.3145 Acc: 0.8730
val Loss: 0.2046 Acc: 0.9216

Epoch 23/24
----------
train Loss: 0.2740 Acc: 0.8689
val Loss: 0.2060 Acc: 0.9216

Epoch 24/24
----------
train Loss: 0.2996 Acc: 0.8607
val Loss: 0.1837 Acc: 0.9281

Training complete in 1m 5s
Best val Acc: 0.941176

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(weights='IMAGENET1K_V1')
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)

Epoch 0/24
----------
train Loss: 0.6996 Acc: 0.6516
val Loss: 0.2014 Acc: 0.9346

Epoch 1/24
----------
train Loss: 0.4233 Acc: 0.8033
val Loss: 0.2656 Acc: 0.8758

Epoch 2/24
----------
train Loss: 0.4603 Acc: 0.7869
val Loss: 0.1847 Acc: 0.9477

Epoch 3/24
----------
train Loss: 0.3096 Acc: 0.8566
val Loss: 0.1747 Acc: 0.9477

Epoch 4/24
----------
train Loss: 0.4427 Acc: 0.8156
val Loss: 0.1630 Acc: 0.9477

Epoch 5/24
----------
train Loss: 0.5505 Acc: 0.7828
val Loss: 0.1643 Acc: 0.9477

Epoch 6/24
----------
train Loss: 0.3004 Acc: 0.8607
val Loss: 0.1744 Acc: 0.9542

Epoch 7/24
----------
train Loss: 0.4083 Acc: 0.8361
val Loss: 0.1892 Acc: 0.9412

Epoch 8/24
----------
train Loss: 0.4483 Acc: 0.7910
val Loss: 0.1984 Acc: 0.9477

Epoch 9/24
----------
train Loss: 0.3335 Acc: 0.8279
val Loss: 0.1942 Acc: 0.9412

Epoch 10/24
----------
train Loss: 0.2413 Acc: 0.8934
val Loss: 0.2001 Acc: 0.9477

Epoch 11/24
----------
train Loss: 0.3107 Acc: 0.8689
val Loss: 0.1801 Acc: 0.9412

Epoch 12/24
----------
train Loss: 0.3032 Acc: 0.8689
val Loss: 0.1669 Acc: 0.9477

Epoch 13/24
----------
train Loss: 0.3587 Acc: 0.8525
val Loss: 0.1900 Acc: 0.9477

Epoch 14/24
----------
train Loss: 0.2771 Acc: 0.8893
val Loss: 0.2317 Acc: 0.9216

Epoch 15/24
----------
train Loss: 0.3064 Acc: 0.8852
val Loss: 0.1909 Acc: 0.9477

Epoch 16/24
----------
train Loss: 0.4243 Acc: 0.8238
val Loss: 0.2227 Acc: 0.9346

Epoch 17/24
----------
train Loss: 0.3297 Acc: 0.8238
val Loss: 0.1916 Acc: 0.9412

Epoch 18/24
----------
train Loss: 0.4235 Acc: 0.8238
val Loss: 0.1766 Acc: 0.9477

Epoch 19/24
----------
train Loss: 0.2500 Acc: 0.8934
val Loss: 0.2003 Acc: 0.9477

Epoch 20/24
----------
train Loss: 0.2413 Acc: 0.8934
val Loss: 0.1821 Acc: 0.9477

Epoch 21/24
----------
train Loss: 0.3762 Acc: 0.8115
val Loss: 0.1842 Acc: 0.9412

Epoch 22/24
----------
train Loss: 0.3485 Acc: 0.8566
val Loss: 0.2166 Acc: 0.9281

Epoch 23/24
----------
train Loss: 0.3625 Acc: 0.8361
val Loss: 0.1747 Acc: 0.9412

Epoch 24/24
----------
train Loss: 0.3840 Acc: 0.8320
val Loss: 0.1768 Acc: 0.9412

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

visualize_model(model_conv)

plt.ioff()
plt.show()

Inference on custom images

Use the trained model to make predictions on custom images and visualize the predicted class labels along with the images.

def visualize_model_predictions(model,img_path):
    was_training = model.training
    model.eval()

    img = Image.open(img_path)
    img = data_transforms['val'](img)
    img = img.unsqueeze(0)
    img = img.to(device)

    with torch.no_grad():
        outputs = model(img)
        _, preds = torch.max(outputs, 1)

        ax = plt.subplot(2,2,1)
        ax.axis('off')
        ax.set_title(f'Predicted: {class_names[preds[0]]}')
        imshow(img.cpu().data[0])

        model.train(mode=was_training)

visualize_model_predictions(
    model_conv,
    img_path='data/hymenoptera_data/val/bees/72100438_73de9f17af.jpg'
)

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.