View the runnable example on GitHub

Accelerate Computer Vision Data Processing Pipeline

You can use transforms and datasets from bigdl.nano.pytorch.vision as a drop-in replacement of torchvision.transforms and torchvision.datasets to easily accelerate your computer vision data processing pipeline in PyTorch Lightning applications.

📝 Note

Before starting your PyTorch Lightning application, it is highly recommended to run source bigdl-nano-init to set several environment variables based on your current hardware. Empirically, these variables will bring big performance increase for most PyTorch Lightning applications on training workloads.

Let’s take a self-defined LightningModule (based on a ResNet-18 model pretrained on ImageNet dataset) as an example, and suppose we would like to finetune the model on OxfordIIITPet dataset:

[ ]:

model = MyLightningModule()

      The definition of MyLightningModule can be found in the runnable example.

To finetune the model on OxfordIIITPet dataset, we need to create required train/validate datasets and dataloaders. To accelerate the data processing pipeline, you could simply import BigDL-Nano transforms and datasets to replace torchvision.transforms and torchvision.datasets:

[ ]:

# from torchvision import transforms
# from torchvision.datasets import OxfordIIITPet
from bigdl.nano.pytorch.vision import transforms
from bigdl.nano.pytorch.vision.datasets import OxfordIIITPet

# Data processing steps are the same as using torchvision
train_transform = transforms.Compose([transforms.Resize(256),
                                        transforms.RandomCrop(224),
                                        transforms.RandomHorizontalFlip(),
                                        transforms.ColorJitter(brightness=.5, hue=.3),
                                        transforms.ToTensor(),
                                        transforms.Normalize([0.485, 0.456, 0.406],
                                                            [0.229, 0.224, 0.225])])
val_transform = transforms.Compose([transforms.Resize(256),
                                    transforms.CenterCrop(224),
                                    transforms.ToTensor(),
                                    transforms.Normalize([0.485, 0.456, 0.406],
                                                            [0.229, 0.224, 0.225])])

train_dataset = OxfordIIITPet(root="/tmp/data", transform=train_transform, download=True)
val_dataset = OxfordIIITPet(root="/tmp/data", transform=val_transform)

[ ]:

train_loader, val_loader = create_dataloaders(train_dataset, val_dataset)

      The definition of create_dataloaders can be found in the runnable example.

You could then do the training and evaluation steps with Nano Trainer:

[ ]:

from bigdl.nano.pytorch import Trainer

trainer = Trainer(max_epochs=5)
trainer.fit(model, train_dataloaders=train_loader)
trainer.validate(model, dataloaders=val_loader)

📚 Related Readings