Ported from Transfer Learning for Computer Vision Tutorial by Sasank Chilamkurthy
Available under the same license (BSD-3-Clause)
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
require "torch"
require "torchvision"
require "numo/narray"
require "matplotlib/iruby"
Matplotlib::IRuby.activate
plt = Matplotlib::Pyplot
plt.ion # interactive modeWe will use TorchVision and Torch::Utils::Data modules 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" => TorchVision::Transforms::Compose.new([
TorchVision::Transforms::RandomResizedCrop.new(224),
TorchVision::Transforms::RandomHorizontalFlip.new,
TorchVision::Transforms::ToTensor.new,
TorchVision::Transforms::Normalize.new([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
]),
"val" => TorchVision::Transforms::Compose.new([
TorchVision::Transforms::Resize.new(256),
TorchVision::Transforms::CenterCrop.new(224),
TorchVision::Transforms::ToTensor.new,
TorchVision::Transforms::Normalize.new([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])
}
data_dir = "hymenoptera_data"
image_datasets =
["train", "val"].to_h do |x|
[x, TorchVision::Datasets::ImageFolder.new(File.join(data_dir, x), transform: data_transforms[x])]
end
dataloaders =
["train", "val"].to_h do |x|
[x, Torch::Utils::Data::DataLoader.new(image_datasets[x], batch_size: 4, shuffle: true)]
end
dataset_sizes = ["train", "val"].to_h { |x| [x, image_datasets[x].size] }
class_names = image_datasets["train"].classes
device = Torch.device(Torch::CUDA.available? ? "cuda:0" : "cpu")Let’s visualize a few training images so as to understand the data augmentations.
def imshow(inp, title: nil)
inp = inp.numo.transpose(1, 2, 0)
mean = Numo::SFloat.cast([0.485, 0.456, 0.406])
std = Numo::SFloat.cast([0.229, 0.224, 0.225])
inp = std * inp + mean
inp = inp.clip(0, 1)
plt = Matplotlib::Pyplot
plt.imshow(inp.to_a)
plt.title(title) if title
plt.pause(0.001) # pause a bit so that plots are updated
end
# Get a batch of training data
inputs, classes = dataloaders["train"].first
# Make a grid from batch
out = TorchVision::Utils.make_grid(inputs)
imshow(out, title: classes.to_a.map { |x| class_names[x] })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::LRScheduler.
train_model = lambda do |model, criterion, optimizer, scheduler, num_epochs=25|
since = Time.now
best_model_wts = model.state_dict.transform_values { |v| v.data.clone }
best_acc = 0.0
num_epochs.times do |epoch|
puts "Epoch #{epoch}/#{num_epochs - 1}"
puts "-" * 10
# Each epoch has a training and validation phase
["train", "val"].each do |phase|
if phase == "train"
model.train # Set model to training mode
else
model.eval # Set model to evaluate mode
end
running_loss = 0.0
running_corrects = 0
# Iterate over data.
dataloaders[phase].each do |inputs, labels|
inputs = inputs.to(device)
labels = labels.to(device)
# zero the parameter gradients
optimizer.zero_grad
# forward
# track history if only in train
loss = nil
preds = nil
Torch.set_grad_enabled(phase == "train") do
outputs = model.call(inputs)
_, preds = Torch.max(outputs, 1)
loss = criterion.call(outputs, labels)
# backward + optimize only if in training phase
if phase == "train"
loss.backward
optimizer.step
end
end
# statistics
running_loss += loss.item * inputs.size(0)
running_corrects += Torch.sum(preds.eq(labels.data)).item
end
if phase == "train"
scheduler.step
end
epoch_loss = running_loss / dataset_sizes[phase]
epoch_acc = running_corrects.to_f / dataset_sizes[phase]
puts "%s Loss: %.4f Acc: %.4f" % [phase, epoch_loss, epoch_acc]
# deep copy the model
if phase == "val" && epoch_acc > best_acc
best_acc = epoch_acc
best_model_wts = model.state_dict.transform_values { |v| v.data.clone }
end
end
puts
end
time_elapsed = Time.now - since
puts "Training complete in %.0fm %.0fs" % [time_elapsed.div(60), time_elapsed % 60]
puts "Best val Acc: %4f" % best_acc
# load best model weights
model.load_state_dict(best_model_wts)
model
endGeneric method to display predictions for a few images
visualize_model = lambda do |model, num_images=6|
# TODO use model.training after 0.8.2 release
was_training = model.instance_variable_get(:@training)
model.eval
images_so_far = 0
fig = plt.figure
Torch.no_grad do
dataloaders["val"].each_with_index do |(inputs, labels), i|
inputs = inputs.to(device)
labels = labels.to(device)
outputs = model.call(inputs)
_, preds = Torch.max(outputs, 1)
inputs.size[0].times do |j|
images_so_far += 1
ax = plt.subplot(num_images.div(2), 2, images_so_far)
ax.axis("off")
ax.set_title("predicted: #{class_names[preds[j].item]}")
imshow(inputs.cpu.data[j])
if images_so_far == num_images
model.train(mode: was_training)
return
end
end
end
model.train(mode: was_training)
end
endUse this script to download the pretrained model.
pip install torchvision
python pretrained.pyLoad a pretrained model and reset final fully connected layer.
model_ft = TorchVision::Models::ResNet18.new
model_ft.load_state_dict(Torch.load("net.pth"))
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 = Torch::NN::Linear.new(num_ftrs, 2)
model_ft = model_ft.to(device)
criterion = Torch::NN::CrossEntropyLoss.new
# Observe that all parameters are being optimized
optimizer_ft = Torch::Optim::SGD.new(model_ft.parameters, lr: 0.001, momentum: 0.9)
# Decay LR by a factor of 0.1 every 7 epochs
exp_lr_scheduler = Torch::Optim::LRScheduler::StepLR.new(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.call(model_ft, criterion, optimizer_ft, exp_lr_scheduler, 25)visualize_model.call(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.
model_conv = TorchVision::Models::ResNet18.new
model_conv.load_state_dict(Torch.load("net.pth"))
model_conv.parameters.each do |param|
param.requires_grad = false
end
# Parameters of newly constructed modules have requires_grad=True by default
num_ftrs = model_conv.fc.in_features
model_conv.fc = Torch::NN::Linear.new(num_ftrs, 2)
model_conv = model_conv.to(device)
criterion = Torch::NN::CrossEntropyLoss.new
# Observe that only parameters of final layer are being optimized as
# opposed to before.
optimizer_conv = Torch::Optim::SGD.new(model_conv.fc.parameters, lr: 0.001, momentum: 0.9)
# Decay LR by a factor of 0.1 every 7 epochs
exp_lr_scheduler = Torch::Optim::LRScheduler::StepLR.new(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.call(model_conv, criterion, optimizer_conv, exp_lr_scheduler, 25)visualize_model.call(model_conv)
plt.ioff
plt.show