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 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 modeWe 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")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])
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':
scheduler.step()
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)
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 modelGeneric 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)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)
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)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: 0.7565 Acc: 0.6598
val Loss: 0.2146 Acc: 0.9085
Epoch 1/24
----------
train Loss: 0.4915 Acc: 0.7951
val Loss: 0.3471 Acc: 0.8889
Epoch 2/24
----------
train Loss: 0.7898 Acc: 0.7541
val Loss: 0.4754 Acc: 0.8497
Epoch 3/24
----------
train Loss: 0.7151 Acc: 0.7295
val Loss: 0.5705 Acc: 0.8235
Epoch 4/24
----------
train Loss: 0.8363 Acc: 0.7459
val Loss: 0.2653 Acc: 0.9020
Epoch 5/24
----------
train Loss: 0.6235 Acc: 0.7992
val Loss: 0.4678 Acc: 0.8366
Epoch 6/24
----------
train Loss: 1.0205 Acc: 0.7131
val Loss: 0.5871 Acc: 0.8235
Epoch 7/24
----------
train Loss: 0.4644 Acc: 0.8238
val Loss: 0.2850 Acc: 0.8824
Epoch 8/24
----------
train Loss: 0.3654 Acc: 0.8566
val Loss: 0.2785 Acc: 0.9085
Epoch 9/24
----------
train Loss: 0.3400 Acc: 0.8648
val Loss: 0.2869 Acc: 0.9085
Epoch 10/24
----------
train Loss: 0.2939 Acc: 0.8770
val Loss: 0.2930 Acc: 0.8889
Epoch 11/24
----------
train Loss: 0.3057 Acc: 0.8811
val Loss: 0.2768 Acc: 0.9216
Epoch 12/24
----------
train Loss: 0.3081 Acc: 0.8689
val Loss: 0.3098 Acc: 0.8889
Epoch 13/24
----------
train Loss: 0.3764 Acc: 0.8607
val Loss: 0.2620 Acc: 0.9150
Epoch 14/24
----------
train Loss: 0.3119 Acc: 0.8689
val Loss: 0.2642 Acc: 0.9216
Epoch 15/24
----------
train Loss: 0.2269 Acc: 0.9180
val Loss: 0.2648 Acc: 0.9281
Epoch 16/24
----------
train Loss: 0.3055 Acc: 0.8893
val Loss: 0.2605 Acc: 0.9281
Epoch 17/24
----------
train Loss: 0.3213 Acc: 0.8730
val Loss: 0.2535 Acc: 0.9216
Epoch 18/24
----------
train Loss: 0.3325 Acc: 0.8566
val Loss: 0.2747 Acc: 0.9281
Epoch 19/24
----------
train Loss: 0.3007 Acc: 0.8648
val Loss: 0.2759 Acc: 0.9150
Epoch 20/24
----------
train Loss: 0.3498 Acc: 0.8443
val Loss: 0.2742 Acc: 0.9216
Epoch 21/24
----------
train Loss: 0.3433 Acc: 0.8443
val Loss: 0.2605 Acc: 0.8954
Epoch 22/24
----------
train Loss: 0.2822 Acc: 0.8689
val Loss: 0.2610 Acc: 0.9281
Epoch 23/24
----------
train Loss: 0.3025 Acc: 0.8648
val Loss: 0.2766 Acc: 0.9150
Epoch 24/24
----------
train Loss: 0.3401 Acc: 0.8607
val Loss: 0.2650 Acc: 0.9085
Training complete in 1m 13s
Best val Acc: 0.928105visualize_model(model_ft)
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)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: 0.6321 Acc: 0.6270
val Loss: 0.2507 Acc: 0.9085
Epoch 1/24
----------
train Loss: 0.6750 Acc: 0.7131
val Loss: 0.1764 Acc: 0.9412
Epoch 2/24
----------
train Loss: 0.4614 Acc: 0.8033
val Loss: 0.1683 Acc: 0.9477
Epoch 3/24
----------
train Loss: 0.5286 Acc: 0.7705
val Loss: 0.1566 Acc: 0.9542
Epoch 4/24
----------
train Loss: 0.3432 Acc: 0.8566
val Loss: 0.1878 Acc: 0.9281
Epoch 5/24
----------
train Loss: 0.4155 Acc: 0.8402
val Loss: 0.1631 Acc: 0.9412
Epoch 6/24
----------
train Loss: 0.4328 Acc: 0.7951
val Loss: 0.1659 Acc: 0.9542
Epoch 7/24
----------
train Loss: 0.3537 Acc: 0.8566
val Loss: 0.1803 Acc: 0.9412
Epoch 8/24
----------
train Loss: 0.3448 Acc: 0.8279
val Loss: 0.1681 Acc: 0.9412
Epoch 9/24
----------
train Loss: 0.3830 Acc: 0.8402
val Loss: 0.1707 Acc: 0.9412
Epoch 10/24
----------
train Loss: 0.2958 Acc: 0.8689
val Loss: 0.1632 Acc: 0.9542
Epoch 11/24
----------
train Loss: 0.2552 Acc: 0.8893
val Loss: 0.1776 Acc: 0.9346
Epoch 12/24
----------
train Loss: 0.2846 Acc: 0.8689
val Loss: 0.1792 Acc: 0.9346
Epoch 13/24
----------
train Loss: 0.3263 Acc: 0.8566
val Loss: 0.1695 Acc: 0.9412
Epoch 14/24
----------
train Loss: 0.3925 Acc: 0.8361
val Loss: 0.2126 Acc: 0.9216
Epoch 15/24
----------
train Loss: 0.3780 Acc: 0.8074
val Loss: 0.1637 Acc: 0.9477
Epoch 16/24
----------
train Loss: 0.3619 Acc: 0.8525
val Loss: 0.1704 Acc: 0.9412
Epoch 17/24
----------
train Loss: 0.2966 Acc: 0.8689
val Loss: 0.1672 Acc: 0.9346
Epoch 18/24
----------
train Loss: 0.2811 Acc: 0.8934
val Loss: 0.1799 Acc: 0.9412
Epoch 19/24
----------
train Loss: 0.2484 Acc: 0.8893
val Loss: 0.1859 Acc: 0.9281
Epoch 20/24
----------
train Loss: 0.2696 Acc: 0.8975
val Loss: 0.1903 Acc: 0.9412
Epoch 21/24
----------
train Loss: 0.3679 Acc: 0.8197
val Loss: 0.1866 Acc: 0.9412
Epoch 22/24
----------
train Loss: 0.2854 Acc: 0.8852
val Loss: 0.1814 Acc: 0.9346
Epoch 23/24
----------
train Loss: 0.3550 Acc: 0.8443
val Loss: 0.1677 Acc: 0.9412
Epoch 24/24
----------
train Loss: 0.3928 Acc: 0.8033
val Loss: 0.1571 Acc: 0.9477
Training complete in 0m 34s
Best val Acc: 0.954248visualize_model(model_conv)
plt.ioff()
plt.show()
Total running time of the script: ( 1 minutes 58.873 seconds)
Download Python source code: transfer_learning_tutorial.pyDownload Jupyter notebook: transfer_learning_tutorial.ipynb