Transfer Learning tutorial

Author: Sasank Chilamkurthy

In this tutorial, you will learn how to train your network using transfer learning. You can read more about the transfer learning at cs231n notes

Quoting this 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 looks 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.autograd import Variable
import numpy as np
import torchvision
from torchvision import datasets, models, transforms
import matplotlib.pyplot as plt
import time
import copy
import os

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.RandomSizedCrop(224),
        transforms.RandomHorizontalFlip(),
        transforms.ToTensor(),
        transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
    ]),
    'val': transforms.Compose([
        transforms.Scale(256),
        transforms.CenterCrop(224),
        transforms.ToTensor(),
        transforms.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
    ]),
}

data_dir = 'hymenoptera_data'
dsets = {x: datasets.ImageFolder(os.path.join(data_dir, x), data_transforms[x])
         for x in ['train', 'val']}
dset_loaders = {x: torch.utils.data.DataLoader(dsets[x], batch_size=4,
                                               shuffle=True, num_workers=4)
                for x in ['train', 'val']}
dset_sizes = {x: len(dsets[x]) for x in ['train', 'val']}
dset_classes = dsets['train'].classes

use_gpu = torch.cuda.is_available()

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
    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(dset_loaders['train']))

# Make a grid from batch
out = torchvision.utils.make_grid(inputs)

imshow(out, title=[dset_classes[x] for x in classes])
../_images/sphx_glr_transfer_learning_tutorial_001.png

Training the model

Now, let’s write a general function to train a model. Here, we will illustrate:

  • Scheduling the learning rate
  • Saving (deep copying) the best model

In the following, parameter lr_scheduler(optimizer, epoch) is a function which modifies optimizer so that the learning rate is changed according to desired schedule.

def train_model(model, criterion, optimizer, lr_scheduler, num_epochs=25):
    since = time.time()

    best_model = model
    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':
                optimizer = lr_scheduler(optimizer, epoch)
                model.train(True)  # Set model to training mode
            else:
                model.train(False)  # Set model to evaluate mode

            running_loss = 0.0
            running_corrects = 0

            # Iterate over data.
            for data in dset_loaders[phase]:
                # get the inputs
                inputs, labels = data

                # wrap them in Variable
                if use_gpu:
                    inputs, labels = Variable(inputs.cuda()), \
                        Variable(labels.cuda())
                else:
                    inputs, labels = Variable(inputs), Variable(labels)

                # zero the parameter gradients
                optimizer.zero_grad()

                # forward
                outputs = model(inputs)
                _, preds = torch.max(outputs.data, 1)
                loss = criterion(outputs, labels)

                # backward + optimize only if in training phase
                if phase == 'train':
                    loss.backward()
                    optimizer.step()

                # statistics
                running_loss += loss.data[0]
                running_corrects += torch.sum(preds == labels.data)

            epoch_loss = running_loss / dset_sizes[phase]
            epoch_acc = running_corrects / dset_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 = copy.deepcopy(model)

        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))
    return best_model

Learning rate scheduler

Let’s create our learning rate scheduler. We will exponentially decrease the learning rate once every few epochs.

def exp_lr_scheduler(optimizer, epoch, init_lr=0.001, lr_decay_epoch=7):
    """Decay learning rate by a factor of 0.1 every lr_decay_epoch epochs."""
    lr = init_lr * (0.1**(epoch // lr_decay_epoch))

    if epoch % lr_decay_epoch == 0:
        print('LR is set to {}'.format(lr))

    for param_group in optimizer.param_groups:
        param_group['lr'] = lr

    return optimizer

Visualizing the model predictions

Generic function to display predictions for a few images

def visualize_model(model, num_images=6):
    images_so_far = 0
    fig = plt.figure()

    for i, data in enumerate(dset_loaders['val']):
        inputs, labels = data
        if use_gpu:
            inputs, labels = Variable(inputs.cuda()), Variable(labels.cuda())
        else:
            inputs, labels = Variable(inputs), Variable(labels)

        outputs = model(inputs)
        _, preds = torch.max(outputs.data, 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(dset_classes[labels.data[j]]))
            imshow(inputs.cpu().data[j])

            if images_so_far == num_images:
                return

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
model_ft.fc = nn.Linear(num_ftrs, 2)

if use_gpu:
    model_ft = model_ft.cuda()

criterion = nn.CrossEntropyLoss()

# Observe that all parameters are being optimized
optimizer_ft = optim.SGD(model_ft.parameters(), lr=0.001, momentum=0.9)

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
----------
LR is set to 0.001
train Loss: 0.1565 Acc: 0.6844
val Loss: 0.1242 Acc: 0.7908

Epoch 1/24
----------
train Loss: 0.1269 Acc: 0.7582
val Loss: 0.0473 Acc: 0.9150

Epoch 2/24
----------
train Loss: 0.1062 Acc: 0.8156
val Loss: 0.0714 Acc: 0.8562

Epoch 3/24
----------
train Loss: 0.0993 Acc: 0.8156
val Loss: 0.0566 Acc: 0.9085

Epoch 4/24
----------
train Loss: 0.0962 Acc: 0.8730
val Loss: 0.2831 Acc: 0.6863

Epoch 5/24
----------
train Loss: 0.1778 Acc: 0.7787
val Loss: 0.4696 Acc: 0.6667

Epoch 6/24
----------
train Loss: 0.1765 Acc: 0.8238
val Loss: 0.0994 Acc: 0.8758

Epoch 7/24
----------
LR is set to 0.0001
train Loss: 0.1600 Acc: 0.8156
val Loss: 0.0850 Acc: 0.8824

Epoch 8/24
----------
train Loss: 0.0876 Acc: 0.8730
val Loss: 0.0841 Acc: 0.8889

Epoch 9/24
----------
train Loss: 0.1105 Acc: 0.8279
val Loss: 0.0777 Acc: 0.8954

Epoch 10/24
----------
train Loss: 0.0976 Acc: 0.8607
val Loss: 0.0719 Acc: 0.9020

Epoch 11/24
----------
train Loss: 0.0644 Acc: 0.9057
val Loss: 0.0637 Acc: 0.9085

Epoch 12/24
----------
train Loss: 0.1003 Acc: 0.8279
val Loss: 0.0678 Acc: 0.9085

Epoch 13/24
----------
train Loss: 0.0755 Acc: 0.8770
val Loss: 0.0576 Acc: 0.9150

Epoch 14/24
----------
LR is set to 1.0000000000000003e-05
train Loss: 0.0642 Acc: 0.9139
val Loss: 0.0608 Acc: 0.9150

Epoch 15/24
----------
train Loss: 0.0870 Acc: 0.8484
val Loss: 0.0618 Acc: 0.9085

Epoch 16/24
----------
train Loss: 0.0704 Acc: 0.8975
val Loss: 0.0629 Acc: 0.9085

Epoch 17/24
----------
train Loss: 0.0714 Acc: 0.8730
val Loss: 0.0627 Acc: 0.9150

Epoch 18/24
----------
train Loss: 0.0739 Acc: 0.8730
val Loss: 0.0637 Acc: 0.9020

Epoch 19/24
----------
train Loss: 0.0865 Acc: 0.8320
val Loss: 0.0701 Acc: 0.8954

Epoch 20/24
----------
train Loss: 0.0868 Acc: 0.8566
val Loss: 0.0654 Acc: 0.9216

Epoch 21/24
----------
LR is set to 1.0000000000000002e-06
train Loss: 0.0868 Acc: 0.8525
val Loss: 0.0619 Acc: 0.9216

Epoch 22/24
----------
train Loss: 0.0867 Acc: 0.8730
val Loss: 0.0742 Acc: 0.8824

Epoch 23/24
----------
train Loss: 0.0860 Acc: 0.8566
val Loss: 0.0603 Acc: 0.9150

Epoch 24/24
----------
train Loss: 0.1004 Acc: 0.8402
val Loss: 0.0652 Acc: 0.9085

Training complete in 1m 8s
Best val Acc: 0.921569
visualize_model(model_ft)
../_images/sphx_glr_transfer_learning_tutorial_002.png

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)

if use_gpu:
    model_conv = model_conv.cuda()

criterion = nn.CrossEntropyLoss()

# Observe that only parameters of final layer are being optimized as
# opoosed to before.
optimizer_conv = optim.SGD(model_conv.fc.parameters(), lr=0.001, momentum=0.9)

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
----------
LR is set to 0.001
train Loss: 0.1731 Acc: 0.6025
val Loss: 0.0868 Acc: 0.8627

Epoch 1/24
----------
train Loss: 0.1219 Acc: 0.7336
val Loss: 0.0470 Acc: 0.9542

Epoch 2/24
----------
train Loss: 0.1205 Acc: 0.8033
val Loss: 0.0505 Acc: 0.9412

Epoch 3/24
----------
train Loss: 0.0967 Acc: 0.8156
val Loss: 0.0736 Acc: 0.8954

Epoch 4/24
----------
train Loss: 0.1452 Acc: 0.7664
val Loss: 0.0447 Acc: 0.9346

Epoch 5/24
----------
train Loss: 0.1170 Acc: 0.7992
val Loss: 0.0791 Acc: 0.8758

Epoch 6/24
----------
train Loss: 0.1260 Acc: 0.7787
val Loss: 0.0551 Acc: 0.9346

Epoch 7/24
----------
LR is set to 0.0001
train Loss: 0.0580 Acc: 0.9016
val Loss: 0.0473 Acc: 0.9412

Epoch 8/24
----------
train Loss: 0.0950 Acc: 0.8115
val Loss: 0.0579 Acc: 0.9216

Epoch 9/24
----------
train Loss: 0.0868 Acc: 0.8689
val Loss: 0.0500 Acc: 0.9477

Epoch 10/24
----------
train Loss: 0.1027 Acc: 0.8033
val Loss: 0.0476 Acc: 0.9412

Epoch 11/24
----------
train Loss: 0.0911 Acc: 0.8443
val Loss: 0.0500 Acc: 0.9477

Epoch 12/24
----------
train Loss: 0.0968 Acc: 0.8361
val Loss: 0.0520 Acc: 0.9412

Epoch 13/24
----------
train Loss: 0.0930 Acc: 0.8361
val Loss: 0.0501 Acc: 0.9412

Epoch 14/24
----------
LR is set to 1.0000000000000003e-05
train Loss: 0.1072 Acc: 0.7992
val Loss: 0.0631 Acc: 0.9150

Epoch 15/24
----------
train Loss: 0.1021 Acc: 0.8197
val Loss: 0.0458 Acc: 0.9477

Epoch 16/24
----------
train Loss: 0.1014 Acc: 0.8279
val Loss: 0.0467 Acc: 0.9477

Epoch 17/24
----------
train Loss: 0.0645 Acc: 0.8934
val Loss: 0.0517 Acc: 0.9477

Epoch 18/24
----------
train Loss: 0.0704 Acc: 0.8811
val Loss: 0.0511 Acc: 0.9346

Epoch 19/24
----------
train Loss: 0.0821 Acc: 0.8648
val Loss: 0.0469 Acc: 0.9412

Epoch 20/24
----------
train Loss: 0.0838 Acc: 0.8730
val Loss: 0.0431 Acc: 0.9477

Epoch 21/24
----------
LR is set to 1.0000000000000002e-06
train Loss: 0.0964 Acc: 0.8443
val Loss: 0.0433 Acc: 0.9477

Epoch 22/24
----------
train Loss: 0.0823 Acc: 0.8525
val Loss: 0.0462 Acc: 0.9477

Epoch 23/24
----------
train Loss: 0.0769 Acc: 0.8525
val Loss: 0.0452 Acc: 0.9542

Epoch 24/24
----------
train Loss: 0.0953 Acc: 0.8320
val Loss: 0.0594 Acc: 0.9281

Training complete in 0m 35s
Best val Acc: 0.954248
visualize_model(model_conv)

plt.ioff()
plt.show()
../_images/sphx_glr_transfer_learning_tutorial_003.png

Total running time of the script: ( 1 minutes 48.038 seconds)

Generated by Sphinx-Gallery